Classification of team sport activities using a single wearable tracking device

An investigation of an athlete injury likelihood monitoring system using the random forest algorithm and DWT

BACKGROUND

The main goal of sports science is to monitor sports injuries. Nevertheless, the existing sports injury monitoring projects have many expensive instruments and excessively extended monitoring periods, which makes it difficult to expand sports injury monitoring on a large scale.

OBJECTIVE

The advancement of machine learning algorithms opens up new avenues for the tracking of sports injuries.

METHODS

A training set of sports injuries was created using the Discrete Wavelet Transform (DWT) and Random Forest algorithms. Next, a basic analytic framework was created based on the lower-body movement of runners, and an athlete's injury likelihood monitoring system was established. First off, the wearable gyroscope device can efficiently plot the motion displacement curve and monitor the three-dimensional mechanics of the athlete's hips, thighs, and calves. Secondly, the system has a higher computational efficiency and an advantage over other classifier-based systems in terms of testing and training timesRESULTS: The suggested system framework identifies athletes' injury propensity, providing preventive recommendations based on displacement curves, and offering a low total cost and high testing accuracy, making it easy to implement and cost-effective.

CONCLUSION

All things considered, the sports injury monitoring device is very accurate and reasonably priced, making it appropriate for widespread use.

A Quantitative Assessment Grading Study of Balance Performance Based on Lower Limb Dataset

Balance ability is one of the important factors in measuring human physical fitness and a common index for evaluating sports performance. Its quality directly affects the coordination ability of human movements and plays an important role in human productive activities. In the field of sports, balance ability is an important indicator of athletes' selection and training. How to objectively analyze balance performance becomes a problem for every non-professional sports enthusiast. Therefore, in this paper, we used a dataset of lower limb collected by inertial sensors to extract the feature parameters, then designed a RUS Boost classifier for unbalanced data whose basic classifier was SVM model to predict three classifications of balance degree, and, finally, evaluated the performance of the new classifier by comparing it with two basic classifiers (KNN, SVM). The result showed that the new classifier could be used to evaluate the balanced ability of lower limb, and performed higher than basic ones (RUS Boost: 72%; KNN: 60%; SVM: 44%). The results meant the established classification model could be used for and quantitative assessment of balance ability in initial screening and targeted training.

Applications of regularized regression models in sports biomechanics research

Research in sports biomechanics often relies on the use of ordinary least squares (OLS) regression. However, since sports biomechanics research is often characterised by high-dimensional data sets with many predictor variables and few observations, use of OLS regression can sometimes be problematic from a statistical perspective. Statistical learning methods may provide alternate ways to deal with high-dimensional data sets and partially address these problems. For example, regularisation adds penalties to the cost function of OLS regression models, which shrinks large regression coefficients and decreases the model's sensitivity to noise in the data. Regularised regression models also protect against overfitting, improve generalisability, and can be used for variable selection. A short review of biomechanics research studies illustrates how these models provided ways to reduce the number of variables within a model and select only the primary predictors of performance, which helped with the interpretation of results and identified distinct combinations of key predictors of performance. In addition, we illustrate how these models are applied to two sports biomechanics datasets. Given the advantages, sports biomechanists may want to consider the use of regularised regression models in their research design and statistical analyses. Careful consideration should be given, however, to the construction, validation, and interpretation of these models considering their underpinning assumptions and limitations.

Training and match demands differ between the regular season and finals in semi-professional basketball

Basketball competitions often include a scheduled regular season followed by knock-out finals. Understanding training and match demands through the season can help optimize performance and reduce injury risk. This study investigated whether training and/or match demands differed between the regular season and finals, and whether these differences were dependent on player role. Average session intensity and volume and durations of relative exercise intensities (inactive, light, moderate-vigorous, maximal, supramaximal) were quantified during training sessions and matches using accelerometry in two semi-professional basketball teams (n = 23; 10 women, 13 men). Training and match demands were compared between the regular season (training: 445 observations; matches: 387 observations) and finals (training: 113 observations, matches: 75 observations) with consideration of player role (starters, in-rotation bench, out-rotation bench). During finals matches, starters received 4.4 min more playing time (p = 0.03), performed 14% more absolute maximal activity (p < 0.01) and had 8% less relative inactive time (p = 0.02) when compared to the regular season. Out-rotation bench players received 2.1 min less playing time (p < 0.01), performed 33% less absolute maximal activity (p = 0.01) and 57% less absolute supramaximal activity (p < 0.01) in finals when compared to the regular season. During finals training sessions, average training intensity was 5% higher (p = 0.02), absolute moderate-vigorous activity was 3% higher (p = 0.04), relative maximal activity was 12% higher (p < 0.01), and relative inactive time was 5% lower (p = 0.03) when compared to the regular season. These findings suggest starters need to be physically prepared for greater match demands during finals, while out-rotation bench players should supplement their training during finals with extra supramaximal activity to maintain their conditioning levels for matches.

On-Court Activity and Game-Related Statistics during Scoring Streaks in Basketball: Applied Use of Accelerometers

The aim of this observational study was to determine if on-court activity and match statistics differed between periods of scoring streaks and regular play in basketball. Thirty-seven basketballers including professional women, semi-professional women and semi-professional men wore accelerometers during competitive matches throughout a season. Accelerometry-derived live-time individual on-court exercise intensity and team game-related statistics were compared between scoring streaks (periods of play where the teams participating in the study scored at least three times in a row), streaks against (periods of play where the opposition teams scored at least three times in a row) and regular play. Few differences existed in the average exercise intensity between streak types. During streaks against, there was a 5–15% lower proportion of 2-point attempts, 0.8–1.3 fewer defensive rebounds per minute and 0.3–1.6 fewer shot attempts per minute compared to regular play and scoring streaks, and there were 0.3 fewer offensive rebounds per minute compared to regular play. During scoring streaks, there were 0.5 more defensive rebounds per minute, 1.3 more shot attempts per minute, a 43% greater shooting percentage and a 10% lower proportion of 3-point attempts compared to regular play. To reduce the chances of streaks against, teams should focus on facilitating 2-point shot attempts and consider implementing a 3:1 ratio of 2-point to 3-point attempts to maximize scoring success, and they should focus on winning rebounds to facilitate more shot attempts.

Tracking Systems in Team Sports: A Narrative Review of Applications of the Data and Sport Specific Analysis

Seeking to obtain a competitive advantage and manage the risk of injury, team sport organisations are investing in tracking systems that can quantify training and competition characteristics. It is expected that such information can support objective decision-making for the prescription and manipulation of training load. This narrative review aims to summarise, and critically evaluate, different tracking systems and their use within team sports. The selection of systems should be dependent upon the context of the sport and needs careful consideration by practitioners. The selection of metrics requires a critical process to be able to describe, plan, monitor and evaluate training and competition characteristics of each sport. An emerging consideration for tracking systems data is the selection of suitable time analysis, such as temporal durations, peak demands or time series segmentation, whose best use depends on the temporal characteristics of the sport. Finally, examples of characteristics and the application of tracking data across seven popular team sports are presented. Practitioners working in specific team sports are advised to follow a critical thinking process, with a healthy dose of scepticism and awareness of appropriate theoretical frameworks, where possible, when creating new or selecting an existing metric to profile team sport athletes.

CARL: a running recognition algorithm for free-living accelerometer data

Wearable accelerometers hold great promise for physical activity epidemiology and sports biomechanics. However, identifying and extracting data from specific physical activities, such as running, remains challenging.Objective. To develop and validate an algorithm to identify bouts of running in raw, free-living accelerometer data from devices worn at the wrist or torso (waist, hip, chest).Approach. The CARL (continuous amplitude running logistic) classifier identifies acceleration data with amplitude and frequency characteristics consistent with running. The CARL classifier was trained on data from 31 adults wearing accelerometers on the waist and wrist, then validated on free-living data from 30 new, unseen subjects plus 166 subjects from previously-published datasets using different devices, wear locations, and sample frequencies.Main results. On free-living data, the CARL classifier achieved mean accuracy (F₁score) of 0.984 (95% confidence interval 0.962-0.996) for data from the waist and 0.994 (95% CI 0.991-0.996) for data from the wrist. In previously-published datasets, the CARL classifier identified running with mean accuracy (F₁score) of 0.861 (95% CI 0.836-0.884) for data from the chest, 0.911 (95% CI 0.884-0.937) for data from the hip, 0.916 (95% CI 0.877-0.948) for data from the waist, and 0.870 (95% CI 0.834-0.903) for data from the wrist. Misclassification primarily occurred during activities with similar torso acceleration profiles to running, such as rope jumping and elliptical machine use.Significance. The CARL classifier can accurately identify bouts of running as short as three seconds in free-living accelerometry data. An open-source implementation of the CARL classifier is available atgithub.com/johnjdavisiv/carl.

Criterion Validity of an Automated Method of Detecting Live Play Periods in Basketball

This study aimed to develop an automated method to detect live play periods from accelerometry-derived relative exercise intensity in basketball, and to assess the criterion validity of this method. Relative exercise intensity (% oxygen uptake reserve) was quantified for two men's semi-professional basketball matches. Live play period durations were automatically determined using a moving average sample window and relative exercise intensity threshold, and manually determined using annotation of video footage. The sample window duration and intensity threshold were optimised to determine the input parameters for the automated method that would result in the most similarity to the manual method. These input parameters were used to compare the automated and manual active play period durations in another men's semi-professional match and a women's professional match to assess the criterion validity of the automated method. The optimal input parameters were a 9-s sample window and relative exercise intensity threshold of 31% oxygen uptake reserve. The automated method showed good relative (ρ = 0.95–0.96 and ICC = 0.96–0.98, p < 0.01) and absolute (median bias = 0 s) agreement with the manual method. These findings support the use of an automated method using accelerometry-derived relative exercise intensity and a moving average sample window to detect live play periods in basketball.

The influence of playing surface on external demands and physiological responses during a soccer match simulation

We investigated the effects of playing surfaces with different impact absorption characteristics on external demand and physiological responses. Fifteen participants completed a soccer match simulation on natural grass, synthetic turf and concrete surfaces. Accelerometry-derived PlayerLoad^TM per minute (PL·min^-1) and average net force (AvF_Net) were used to quantify external demands at the centre of mass (CoM), upper-back, mid-back and hip. Heart rate, oxygen uptake, energy expenditure and RPE quantified physiological responses. The concrete surface exhibited the least impact absorption, with peak decelerations ~3.5x synthetic turf and ~10x natural grass (p < 0.001). Despite this, there was no differences in external demand between surfaces (surface: p ≥ 0.194; η²_p≤0.092). Both AvF_Net and PL·min^-1 (location: p < 0.001; η²_p≥0.859) were higher at the hip (613(91)N; 12.5(1.2)arb.u), reduced at the mid-back (521(67)N; 8.8(0.7)arb.u) and upper-back (502(60)N; 8.8(0.7)arb.u) when compared to CoM (576(78)N; 10.7(1.0)arb.u). Although playing surface did not influence the external demands, heart rate or oxygen uptake (p > 0.05), energy expenditure was highest on natural grass compared to synthetic turf (P = 0.034) and RPE was highest on synthetic turf compared to concrete (p = 0.026). Different playing surfaces can alter physiological responses to soccer-specific activity even when the external demands are similar.

Automated Classification of Changes of Direction in Soccer Using Inertial Measurement Units

Changes of direction (COD) are an important aspect of soccer match play. Understanding the physiological and biomechanical demands on players in games allows sports scientists to effectively train and rehabilitate soccer players. COD are conventionally recorded using manually annotated time-motion video analysis which is highly time consuming, so more time-efficient approaches are required. The aim was to develop an automated classification model based on multi-sensor player tracking device data to detect COD > 45°. Video analysis data and individual multi-sensor player tracking data (GPS, accelerometer, gyroscopic) for 23 academy-level soccer players were used. A novel 'GPS-COD Angle' variable was developed and used in model training; along with 24 GPS-derived, gyroscope and accelerometer variables. Video annotation was the ground truth indicator of occurrence of COD > 45°. The random forest classifier using the full set of features demonstrated the highest accuracy (AUROC = 0.957, 95% CI = 0.956-0.958, Sensitivity = 0.941, Specificity = 0.772. To balance sensitivity and specificity, model parameters were optimised resulting in a value of 0.889 for both metrics. Similarly high levels of accuracy were observed for random forest models trained using a reduced set of features, accelerometer-derived variables only, and gyroscope-derived variables only. These results point to the potential effectiveness of the novel methodology implemented in automatically identifying COD in soccer players.

A Wearable System for the Estimation of Performance-Related Metrics during Running and Jumping Tasks

Athletic performance, technique assessment, and injury prevention are all important aspects in sports for both professional and amateur athletes. Wearable technology is attracting the research community’s interest because of its capability to provide real-time biofeedback to coaches and athletes when on the field and outside of more restrictive laboratory conditions. In this paper, a novel wearable motion sensor-based system has been designed and developed for athletic performance assessment during running and jumping tasks. The system consists of a number of components involving embedded systems (hardware and software), back-end analytics, information and communications technology (ICT) platforms, and a graphical user interface for data visualization by the coach. The system is able to provide automatic activity recognition, estimation of running and jumping metrics, as well as vertical ground reaction force (GRF) predictions, with sufficient accuracy to provide valuable information as regards training outcomes. The developed system is low-power, sufficiently small for real-world scenarios, easy to use, and achieves the specified communication range. The system’s high sampling rate, levels of accuracy and performance enables it as a performance evaluation tool able to support coaches and athletes in their real-world practice.

Classification of Australian football kick types in-situation via ankle-mounted inertial measurement units

The utility of inertial measurement units (IMUs) for sporting skill and performance analysis during training and competition is advantageous for enhancing the objectivity of athlete monitoring. This study aimed to classify Australian Rules football (AF) kick types in an applied environment using ankle-mounted IMUs. IMUs and video capture of a controlled protocol, including four kick types at varying distances, were recorded during a single testing session with female AF athletes (n = 20). Processed IMU data were modelled using support vector machine classifier, random forest, and k-nearest neighbour algorithms under a 2-Kick, 4-Kick, and kick distance (10, 20, 30 m) conditions. The random forest model showed the highest results for overall classification accuracy (83% 2-Kick and 80% 4-Kick), test F1-score (0.76 2-Kick and 0.81 4-Kick), and AUC score (0.58 2-Kick and 0.60 4-Kick). Kick distance classification showed a model test and class weighted F1-score of 0.63 and overall accuracy of 64%, respectively. This study highlights the potential for an applied semi-automated AF training kick detection and type classification system using IMUs.

Physical characteristics of different professional rugby union competition levels

OBJECTIVES

To evaluate whether differences in physical characteristics (running-related and collision-related metrics) exist between four different professional rugby union competition levels.

DESIGN

We collected and retrospectively analysed microsensor technology data from players of two professional rugby union clubs that competed across four different competition levels: International rugby union, European Rugby Champions Cup, PRO14 club competition, and British and Irish Cup.

METHODS

Differences between competition levels were analysed using a one-way ANOVA test. The Tukey HSD test was completed to perform multiple pairwise-comparisons between the means of the competition levels and player positional groups.

RESULTS

Ten of the 12 microsensor technology derived physical characteristics were significantly different between competition levels. Collision load-, collisions-, and high metabolic load efforts-per minute all increased at higher competition levels. These differences were also noted across player positional groups.

CONCLUSIONS

The physical characteristics of rugby union match-play differ across competitions levels. Our data suggest that professional rugby union players require specific physical preparation for different competition levels. In particular, players are likely to need specific preparation and recovery for the higher collision intensity observed at higher competition levels.

Effect of Player Role and Competition Level on Player Demands in Basketball

This study compared basketball training and match demands between player roles (starters, in-rotation bench players, out-rotation bench players) and between competition levels (semi-professional, professional). Thirty-seven players from one professional women's team, one semi-professional women's team, and one semi-professional men's team wore accelerometers during training and matches throughout a competitive season. All teams were used for player role comparisons and the women's teams were used to compare competition levels. Match and training session average intensity and volume, and durations of relative exercise intensities (inactive, light, moderate-vigorous, maximal, supramaximal) were calculated. Compared to out-rotation bench players, starters experienced twice the average match intensity and volume, spent 50% less match time being inactive, and spent 1.7-4.2× more match time in all other activity categories (p < 0.01). Compared to in-rotation bench players, starters experienced 1.2× greater average match intensity and volume, spent 17% less match time being inactive, and spent 1.4-1.5× more match time performing moderate-vigorous and maximal activity (p < 0.01). No differences in match demands were found between women's competition levels, however the professional team experienced double the cumulative weekly training volume of the semi-professional team and spent 1.6-2.1× more cumulative weekly time in all activity categories (p < 0.01). To improve performance and reduce injury risk, players should prepare for the greatest match demands they could encounter during a season while considering potential changes to their role. Additionally, players might need their training volume managed when transitioning from a semi-professional to a professional season to reduce the injury risk from sharp increases in training demands.

Wearable Technology and Analytics as a Complementary Toolkit to Optimize Workload and to Reduce Injury Burden

Wearable sensors enable the real-time and non-invasive monitoring of biomechanical, physiological, or biochemical parameters pertinent to the performance of athletes. Sports medicine researchers compile datasets involving a multitude of parameters that can often be time consuming to analyze in order to create value in an expeditious and accurate manner. Machine learning and artificial intelligence models may aid in the clinical decision-making process for sports scientists, team physicians, and athletic trainers in translating the data acquired from wearable sensors to accurately and efficiently make decisions regarding the health, safety, and performance of athletes. This narrative review discusses the application of commercial sensors utilized by sports teams today and the emergence of descriptive analytics to monitor the internal and external workload, hydration status, sleep, cardiovascular health, and return-to-sport status of athletes. This review is written for those who are interested in the application of wearable sensor data and data science to enhance performance and reduce injury burden in athletes of all ages.

Determining Physical Activity Characteristics from Wristband Data for Use in Automated Insulin Delivery Systems

Algorithms that can determine the type of physical activity (PA) and quantify the intensity can allow precision medicine approaches, such as automated insulin delivery systems that modulate insulin administration in response to PA. In this work, data from a multi-sensor wristband is used to design classifiers to distinguish among five different physical states (PS) (resting, activities of daily living, running, biking, and resistance training), and to develop models to estimate the energy expenditure (EE) of the PA for diabetes therapy. The data collected are filtered, features are extracted from the reconciled signals, and the extracted features are used by machine learning algorithms, including deep-learning techniques, to obtain accurate PS classification and EE estimation. The various machine learning techniques have different success rates ranging from 75.7% to 94.8% in classifying the five different PS. The deep neural network model with long short-term memory has 94.8% classification accuracy. We achieved 0.5 MET (Metabolic Equivalent of Task) root-mean-square error for EE estimation accuracy, relative to indirect calorimetry with randomly selected testing data (10% of collected data). We also demonstrate a 5% improvement in PS classification accuracy and a 0.34 MET decrease in the mean absolute error when using multi-sensor approach relative to using only accelerometer data.

Auto detecting deliveries in elite cricket fast bowlers using microsensors and machine learning

Cricket fast bowlers are at a high risk of injury occurrence, which has previously been shown to be correlated to bowling workloads. This study aimed to develop and test an algorithm that can automatically, reliably and accurately detect bowling deliveries. Inertial sensor data from a Catapult OptimEye S5 wearable device was collected from both national and international level fast bowlers (n = 35) in both training and matches, at various intensities. A machine-learning based approach was used to develop the algorithm. Outputs were compared with over 20,000 manually recorded events. A high Matthews correlation coefficient (r = 0.945) showed very good agreement between the automatically detected bowling deliveries and manually recorded ones. The algorithm was found to be both sensitive and specific in training (96.3%, 98.3%) and matches (99.6%, 96.9%), respectively. Rare falsely classified events were typically warm-up deliveries or throws preceded by a run. Inertial sensors data processed by a machine-learning based algorithm provide a valid tool to automatically detect bowling events, whilst also providing the opportunity to look at performance metrics associated with fast bowling. This offers the possibility to better monitor bowling workloads across a range of intensities to mitigate injury risk potential and maximise performance.

Development of a Human Activity Recognition System for Ballet Tasks

Background

Accurate and detailed measurement of a dancer’s training volume is a key requirement to understanding the relationship between a dancer’s pain and training volume. Currently, no system capable of quantifying a dancer’s training volume, with respect to specific movement activities, exists. The application of machine learning models to wearable sensor data for human activity recognition in sport has previously been applied to cricket, tennis and rugby. Thus, the purpose of this study was to develop a human activity recognition system using wearable sensor data to accurately identify key ballet movements (jumping and lifting the leg). Our primary objective was to determine if machine learning can accurately identify key ballet movements during dance training. The secondary objective was to determine the influence of the location and number of sensors on accuracy.

Results

Convolutional neural networks were applied to develop two models for every combination of six sensors (6, 5, 4, 3, etc.) with and without the inclusion of transition movements. At the first level of classification, including data from all sensors, without transitions, the model performed with 97.8% accuracy. The degree of accuracy reduced at the second (83.0%) and third (75.1%) levels of classification. The degree of accuracy reduced with inclusion of transitions, reduction in the number of sensors and various sensor combinations.

Conclusion

The models developed were robust enough to identify jumping and leg lifting tasks in real-world exposures in dancers. The system provides a novel method for measuring dancer training volume through quantification of specific movement tasks. Such a system can be used to further understand the relationship between dancers’ pain and training volume and for athlete monitoring systems. Further, this provides a proof of concept which can be easily translated to other lower limb dominant sporting activities

Quantifying the Collision Dose in Rugby League: A Systematic Review, Meta-analysis, and Critical Analysis

BACKGROUND

Collisions (i.e. tackles, ball carries, and collisions) in the rugby league have the potential to increase injury risk, delay recovery, and influence individual and team performance. Understanding the collision demands of the rugby league may enable practitioners to optimise player health, recovery, and performance.

OBJECTIVE

The aim of this review was to (1) characterise the dose of collisions experienced within senior male rugby league match-play and training, (2) systematically and critically evaluate the methods used to describe the relative and absolute frequency and intensity of collisions, and (3) provide recommendations on collision monitoring.

METHODS

A systematic search of electronic databases (PubMed, SPORTDiscus, Scopus, and Web of Science) using keywords was undertaken. A meta-analysis provided a pooled mean of collision frequency or intensity metrics on comparable data sets from at least two studies.

RESULTS

Forty-three articles addressing the absolute (n) or relative collision frequency (n min^-1) or intensity of senior male rugby league collisions were included. Meta-analysis of video-based studies identified that forwards completed approximately twice the number of tackles per game than backs (n = 24.6 vs 12.8), whilst ball carry frequency remained similar between backs and forwards (n = 11.4 vs 11.2). Variable findings were observed at the subgroup level with a limited number of studies suggesting wide-running forwards, outside backs, and hit-up forwards complete similar ball carries whilst tackling frequency differed. For microtechnology, at the team level, players complete an average of 32.7 collisions per match. Limited data suggested hit-up and wide-running forwards complete the most collisions per match, when compared to adjustables and outside backs. Relative to playing time, forwards (n min^-1 = 0.44) complete a far greater frequency of collision than backs (n min^-1 = 0.16), with data suggesting hit-up forwards undertake more than adjustables, and outside backs. Studies investigating g force intensity zones utilised five unique intensity schemes with zones ranging from 2-3 g to 13-16 g. Given the disparity between device setups and zone classification systems between studies, further analyses were inappropriate. It is recommended that practitioners independently validate microtechnology against video to establish criterion validity.

CONCLUSIONS

Video- and microtechnology-based methods have been utilised to quantify collisions in the rugby league with differential collision profiles observed between forward and back positional groups, and their distinct subgroups. The ball carry demands of forwards and backs were similar, whilst tackle demands were greater for forwards than backs. Microtechnology has been used inconsistently to quantify collision frequency and intensity. Despite widespread popularity, a number of the microtechnology devices have yet to be appropriately validated. Limitations exist in using microtechnology to quantify collision intensity, including the lack of consistency and limited validation. Future directions include application of machine learning approaches to differentiate types of collisions in microtechnology datasets.

Accurate Ambulatory Gait Analysis in Walking and Running Using Machine Learning Models

Wearable sensors have been proposed as alternatives to traditional laboratory equipment for low-cost and portable real-time gait analysis in unconstrained environments. However, the moderate accuracy of these systems currently limits their widespread use. In this paper, we show that support vector regression (SVR) models can be used to extract accurate estimates of fundamental gait parameters (i.e., stride length, velocity, and foot clearance), from custom-engineered instrumented insoles (SportSole) during walking and running tasks. Additionally, these learning-based models are robust to inter-subject variability, thereby making it unnecessary to collect subject-specific training data. Gait analysis was performed in N=14 healthy subjects during two separate sessions, each including 6-minute bouts of treadmill walking and running at different speeds (i.e., 85% and 115% of each subject's preferred speed). Gait metrics were simultaneously measured with the instrumented insoles and with reference laboratory equipment. SVR models yielded excellent intraclass correlation coefficients (ICC) in all the gait parameters analyzed. Percentage mean absolute errors (MAE%) in stride length, velocity, and foot clearance obtained with SVR models were 1.37%±0.49%, 1.23%±0.27%, and 2.08%±0.72% for walking, 2.59%±0.64%, 2.91%±0.85%, and 5.13%±1.52% for running, respectively. These findings provide evidence that machine learning regression is a promising new approach to improve the accuracy of wearable sensors for gait analysis.

Time to go wild: How to conceptualize and measure process dynamics in real teams with high-resolution

Team processes are interdependent activities among team members that transform inputs into outputs, vary over time, and are critical for team effectiveness. Understanding the temporal dynamics of team processes and related team phenomena with a high-resolution lens (i.e., methods with high sampling rates) is particularly challenging when going “into the wild” (i.e., studying teams operating in their full situated context). We review quantitative field studies using high-resolution methods (e.g., video, chat/text data, archival, wearables) and map out the various temporal lenses for studying team dynamics. We synthesize these different lenses and present an integrated temporal framework that is of help in theorizing about team dynamics. We also provide readers with a “how to” guide that summarizes four essential steps along with analytical methods (e.g., sequential and pattern analyses, mixed-methods research, abductive reasoning) that are applicable to the broad scope of high-resolution methods.

Review on Wearable Technology Sensors Used in Consumer Sport Applications

This review paper discusses the trends and projections for wearable technology in the consumer sports sector (excluding professional sport). Analyzing the role of wearable technology for different users and why there is such a need for these devices in everyday lives. It shows how different sensors are influential in delivering a variety of readings that are useful in many ways regarding sport attributes. Wearables are increasing in function, and through integrating technology, users are gathering more data about themselves. The amount of wearable technology available is broad, each having its own role to play in different industries. Inertial measuring unit (IMU) and Global Positioning System (GPS) sensors are predominantly present in sport wearables but can be programmed for different needs. In this review, the differences are displayed to show which sensors are compatible and which ones can evolve sensor technology for sport applications.

Monitoring Student Activities with Smartwatches: On the Academic Performance Enhancement

Motivated by the importance of studying the relationship between habits of students and their academic performance, daily activities of undergraduate participants have been tracked with smartwatches and smartphones. Smartwatches collect data together with an Android application that interacts with the users who provide the labeling of their own activities. The tracked activities include eating, running, sleeping, classroom-session, exam, job, homework, transportation, watching TV-Series, and reading. The collected data were stored in a server for activity recognition with supervised machine learning algorithms. The methodology for the concept proof includes the extraction of features with the discrete wavelet transform from gyroscope and accelerometer signals to improve the classification accuracy. The results of activity recognition with Random Forest were satisfactory (86.9%) and support the relationship between smartwatch sensor signals and daily-living activities of students which opens the possibility for developing future experiments with automatic activity-labeling, and so forth to facilitate activity pattern recognition to propose a recommendation system to enhance the academic performance of each student.

A change point approach to analysing the match activity profiles of team-sport athletes

In team-sport, physical and skilled output is often described via aggregate parameters including total distance and number of skilled involvements. However, the degree to which these output change throughout a team-sport match, as a function of time, is relatively unknown. This study aimed to identify and describe segments of physical and skilled output in team-sport matches with an example in Australian Football. The relationship between the number of change points and level of similarity was also quantified. A binary segmentation algorithm was applied to the velocity time series, collected via wearable sensors, of 37 Australian football players (age: 23 ± 4 years, height: 187 ± 8 cm, mass: 86 ± 9 kg). A change point quotient of between 1 and 15 was used. For these quotients, descriptive statistics, spectral features and a sum of skilled involvements were extracted. Segment similarity for each quotient was evaluated using a random forest model. The strongest classification features in the model were spectral entropy and skewness. Offensive and defensive involvements were the weakest features for classification, suggesting skilled output is dependent on match circumstances. The methodology presented may have application in comparing the specificity of training to matches and designing match rotation strategies.

Validity of a Microsensor-Based Algorithm for Detecting Scrum Events in Rugby Union

PURPOSE

Commercially available microtechnology devices containing accelerometers, gyroscopes, magnetometers, and global positioning technology have been widely used to quantify the demands of rugby union. This study investigated whether data derived from wearable microsensors can be used to develop an algorithm that automatically detects scrum events in rugby union training and match play.

METHODS

Data were collected from 30 elite rugby players wearing a Catapult OptimEye S5 (Catapult Sports, Melbourne, Australia) microtechnology device during a series of competitive matches (n = 46) and training sessions (n = 51). A total of 97 files were required to "train" an algorithm to automatically detect scrum events using random forest machine learning. A further 310 files from training (n = 167) and match-play (n = 143) sessions were used to validate the algorithm's performance.

RESULTS

Across all positions (front row, second row, and back row), the algorithm demonstrated good sensitivity (91%) and specificity (91%) for training and match-play events when the confidence level of the random forest was set to 50%. Generally, the algorithm had better accuracy for match-play events (93.6%) than for training events (87.6%).

CONCLUSIONS

The scrum algorithm was able to accurately detect scrum events for front-row, second-row, and back-row positions. However, for optimal results, practitioners are advised to use the recommended confidence level for each position to limit false positives. Scrum algorithm detection was better with scrums involving ≥5 players and is therefore unlikely to be suitable for scrums involving 3 players (eg, rugby sevens). Additional contact- and collision-detection algorithms are required to fully quantify rugby union demands.

Association Between Preseason Training and Performance in Elite Australian Football

PURPOSE

To examine the association between preseason training variables and subsequent in-season performance in an elite Australian football team.

METHODS

Data from 41 elite male Australian footballers (mean [SD] age = 23.4 [3.1] y, height =188.4 [7.1] cm, and mass = 86.7 [7.9] kg) were collected from 1 Australian Football League (AFL) club. Preseason training data (external load, internal load, fitness testing, and session participation) were collected across the 17-wk preseason phase (6 and 11 wk post-Christmas). Champion Data© Player Rank (CDPR), coaches' ratings, and round 1 selection were used as in-season performance measures. CDPR and coaches' ratings were examined over the entire season, first half of the season, and the first 4 games. Both Pearson and partial (controlling for AFL age) correlations were calculated to assess if any associations existed between preseason training variables and in-season performance measures. A median split was also employed to differentiate between higher- and lower-performing players for each performance measure.

RESULTS

Preseason training activities appeared to have almost no association with performance measured across the entire season and the first half of the season. However, many preseason training variables were significantly linked with performance measured across the first 4 games. Preseason training variables that were measured post-Christmas were the most strongly associated with in-season performance measures. Specifically, total on-field session rating of perceived exertion post-Christmas, a measurement of internal load, displayed the greatest association with performance.

CONCLUSION

Late preseason training (especially on-field match-specific training) is associated with better performance in the early season.

Feature Selection and Comparison of Machine Learning Algorithms in Classification of Grazing and Rumination Behaviour in Sheep

Grazing and ruminating are the most important behaviours for ruminants, as they spend most of their daily time budget performing these. Continuous surveillance of eating behaviour is an important means for monitoring ruminant health, productivity and welfare. However, surveillance performed by human operators is prone to human variance, time-consuming and costly, especially on animals kept at pasture or free-ranging. The use of sensors to automatically acquire data, and software to classify and identify behaviours, offers significant potential in addressing such issues. In this work, data collected from sheep by means of an accelerometer/gyroscope sensor attached to the ear and collar, sampled at 16 Hz, were used to develop classifiers for grazing and ruminating behaviour using various machine learning algorithms: random forest (RF), support vector machine (SVM), k nearest neighbour (kNN) and adaptive boosting (Adaboost). Multiple features extracted from the signals were ranked on their importance for classification. Several performance indicators were considered when comparing classifiers as a function of algorithm used, sensor localisation and number of used features. Random forest yielded the highest overall accuracies: 92% for collar and 91% for ear. Gyroscope-based features were shown to have the greatest relative importance for eating behaviours. The optimum number of feature characteristics to be incorporated into the model was 39, from both ear and collar data. The findings suggest that one can successfully classify eating behaviours in sheep with very high accuracy; this could be used to develop a device for automatic monitoring of feed intake in the sheep sector to monitor health and welfare.

Machine and deep learning for sport-specific movement recognition: a systematic review of model development and performance

Objective assessment of an athlete's performance is of importance in elite sports to facilitate detailed analysis. The implementation of automated detection and recognition of sport-specific movements overcomes the limitations associated with manual performance analysis methods. The object of this study was to systematically review the literature on machine and deep learning for sport-specific movement recognition using inertial measurement unit (IMU) and, or computer vision data inputs. A search of multiple databases was undertaken. Included studies must have investigated a sport-specific movement and analysed via machine or deep learning methods for model development. A total of 52 studies met the inclusion and exclusion criteria. Data pre-processing, processing, model development and evaluation methods varied across the studies. Model development for movement recognition were predominantly undertaken using supervised classification approaches. A kernel form of the Support Vector Machine algorithm was used in 53% of IMU and 50% of vision-based studies. Twelve studies used a deep learning method as a form of Convolutional Neural Network algorithm and one study also adopted a Long Short Term Memory architecture in their model. The adaptation of experimental set-up, data pre-processing, and model development methods are best considered in relation to the characteristics of the targeted sports movement(s).

Physical activity characterization: does one site fit all?

BACKGROUND

It is evident that a growing number of studies advocate a wrist-worn accelerometer for the assessment of patterns of physical activity a priori, yet the veracity of this site rather than any other body-mounted location for its accuracy in classifying activity is hitherto unexplored.

OBJECTIVE

The objective of this review was to identify the relative accuracy with which physical activities can be classified according to accelerometer site and analytical technique.

METHODS

A search of electronic databases was conducted using Web of Science, PubMed and Google Scholar. This review included studies written in the English language, published between database inception and December 2017, which characterized physical activities using a single accelerometer and reported the accuracy of the technique.

RESULTS

A total of 118 articles were initially retrieved. After duplicates were removed and the remaining articles screened, 32 full-text articles were reviewed, resulting in the inclusion of 19 articles that met the eligibility criteria.

CONCLUSION

There is no 'one site fits all' approach to the selection of accelerometer site location or analytical technique. Research design and focus should always inform the most suitable location of attachment, and should be driven by the type of activity being characterized.

Using wearable sensors to classify subject-specific running biomechanical gait patterns based on changes in environmental weather conditions

Running-related overuse injuries can result from a combination of various intrinsic (e.g., gait biomechanics) and extrinsic (e.g., running surface) risk factors. However, it is unknown how changes in environmental weather conditions affect running gait biomechanical patterns since these data cannot be collected in a laboratory setting. Therefore, the purpose of this study was to develop a classification model based on subject-specific changes in biomechanical running patterns across two different environmental weather conditions using data obtained from wearable sensors in real-world environments. Running gait data were recorded during winter and spring sessions, with recorded average air temperatures of -10° C and +6° C, respectively. Classification was performed based on measurements of pelvic drop, ground contact time, braking, vertical oscillation of pelvis, pelvic rotation, and cadence obtained from 66,370 strides (~11,000/runner) from a group of recreational runners. A non-linear and ensemble machine learning algorithm, random forest (RF), was used to classify and compute a heuristic for determining the importance of each variable in the prediction model. To validate the developed subject-specific model, two cross-validation methods (one-against-another and partitioning datasets) were used to obtain experimental mean classification accuracies of 87.18% and 95.42%, respectively, indicating an excellent discriminatory ability of the RF-based model. Additionally, the ranked order of variable importance differed across the individual runners. The results from the RF-based machine-learning algorithm demonstrates that processing gait biomechanical signals from a single wearable sensor can successfully detect changes to an individual's running patterns based on data obtained in real-world environments.

Towards an Efficient One-Class Classifier for Mobile Devices and Wearable Sensors on the Context of Personal Risk Detection

In this work, we present a first step towards an efficient one-class classifier well suited for mobile devices to be implemented as part of a user application coupled with wearable sensors in the context of personal risk detection. We compared one-class Support Vector Machine (ocSVM) and OCKRA (One-Class K-means with Randomly-projected features Algorithm). Both classifiers were tested using four versions of the publicly available PRIDE (Personal RIsk DEtection) dataset. The first version is the original PRIDE dataset, which is based only on time-domain features. We created a second version that is simply an extension of the original dataset with new attributes in the frequency domain. The other two datasets are a subset of these two versions, after a feature selection procedure based on a correlation matrix analysis followed by a Principal Component Analysis. All experiments were focused on the performance of the classifiers as well as on the execution time during the training and classification processes. Therefore, our goal in this work is twofold: we aim at reducing execution time but at the same time maintaining a good classification performance. Our results show that OCKRA achieved on average, 89.1% of Area Under the Curve (AUC) using the full set of features and 83.7% when trained using a subset of them. Furthermore, regarding execution time, OCKRA reports in the best case a 33.1% gain when using a subset of the feature vector, instead of the full set of features. These results are better than those reported by ocSVM, in which case, even though the AUCs are very close to each other, execution times are significantly higher in all cases, for example, more than 20 h versus less than an hour in the worst-case scenario. Having in mind the trade-off between classification performance and efficiency, our results support the choice of OCKRA as our best candidate so far for a mobile implementation where less processing and memory resources are at hand. OCKRA reports a very encouraging speed-up without sacrificing the classifier performance when using the PRIDE dataset based only on time-domain attributes after a feature selection procedure.

Reliability of Wearable Inertial Measurement Units to Measure Physical Activity in Team Handball

PURPOSE

To assess the reliability and sensitivity of commercially available inertial measurement units to measure physical activity in team handball.

METHOD

Twenty-two handball players were instrumented with 2 inertial measurement units (OptimEye S5; Catapult Sports, Melbourne, Australia) taped together. They participated in either a laboratory assessment (n = 10) consisting of 7 team handball-specific tasks or field assessment (n = 12) conducted in 12 training sessions. Variables, including PlayerLoad™ and inertial movement analysis (IMA) magnitude and counts, were extracted from the manufacturers' software. IMA counts were divided into intensity bands of low (1.5-2.5 m·s^-1), medium (2.5-3.5 m·s^-1), high (>3.5 m·s^-1), medium/high (>2.5 m·s^-1), and total (>1.5 m·s^-1). Reliability between devices and sensitivity was established using coefficient of variation (CV) and smallest worthwhile difference (SWD).

RESULTS

Laboratory assessment: IMA magnitude showed a good reliability (CV = 3.1%) in well-controlled tasks. CV increased (4.4-6.7%) in more-complex tasks. Field assessment: Total IMA counts (CV = 1.8% and SWD = 2.5%), PlayerLoad (CV = 0.9% and SWD = 2.1%), and their associated variables (CV = 0.4-1.7%) showed a good reliability, well below the SWD. However, the CV of IMA increased when categorized into intensity bands (2.9-5.6%).

CONCLUSION

The reliability of IMA counts was good when data were displayed as total, high, or medium/high counts. A good reliability for PlayerLoad and associated variables was evident. The CV of the previously mentioned variables was well below the SWD, suggesting that OptimEye's inertial measurement unit and its software are sensitive for use in team handball.

Trends Supporting the In-Field Use of Wearable Inertial Sensors for Sport Performance Evaluation: A Systematic Review

Recent technological developments have led to the production of inexpensive, non-invasive, miniature magneto-inertial sensors, ideal for obtaining sport performance measures during training or competition. This systematic review evaluates current evidence and the future potential of their use in sport performance evaluation. Articles published in English (April 2017) were searched in Web-of-Science, Scopus, Pubmed, and Sport-Discus databases. A keyword search of titles, abstracts and keywords which included studies using accelerometers, gyroscopes and/or magnetometers to analyse sport motor-tasks performed by athletes (excluding risk of injury, physical activity, and energy expenditure) resulted in 2040 papers. Papers and reference list screening led to the selection of 286 studies and 23 reviews. Information on sport, motor-tasks, participants, device characteristics, sensor position and fixing, experimental setting and performance indicators was extracted. The selected papers dealt with motor capacity assessment (51 papers), technique analysis (163), activity classification (19), and physical demands assessment (61). Focus was placed mainly on elite and sub-elite athletes (59%) performing their sport in-field during training (62%) and competition (7%). Measuring movement outdoors created opportunities in winter sports (8%), water sports (16%), team sports (25%), and other outdoor activities (27%). Indications on the reliability of sensor-based performance indicators are provided, together with critical considerations and future trends.

Systematic evaluation of supervised classifiers for fecal microbiota-based prediction of colorectal cancer

Predicting colorectal cancer (CRC) based on fecal microbiota presents a promising method for non-invasive screening of CRC, but the optimization of classification models remains an unaddressed question. The purpose of this study was to systematically evaluate the effectiveness of different supervised machine-learning models in predicting CRC in two independent eastern and western populations. The structures of intestinal microflora in feces in Chinese population (N = 141) were determined by 454 FLX pyrosequencing, and different supervised classifiers were employed to predict CRC based on fecal microbiota operational taxonomic unit (OTUs). As a result, Bayes Net and Random Forest displayed higher accuracies than other algorithms in both populations, although Bayes Net was found with a lower false negative rate than that of Random Forest. Gut microbiota-based prediction was more accurate than the standard fecal occult blood test (FOBT), and the combination of both approaches further improved the prediction accuracy. Moreover, when unclassified OTUs were used as input, the BayesDMNB text algorithm achieved higher accuracy in the Chinese population (AUC=0.994). Taken together, our results suggest that Bayes Net classification model combined with unclassified OTUs may present an accurate method for predicting CRC based on the compositions of gut microbiota.

Validity and reliability of an accelerometer-based player tracking device

This study aimed to determine the intra- and inter-device accuracy and reliability of wearable athletic tracking devices, under controlled laboratory conditions. A total of nineteen portable accelerometers (Catapult OptimEye S5) were mounted to an aluminum bracket, bolted directly to an Unholtz Dickie 20K electrodynamic shaker table, and subjected to a series of oscillations in each of three orthogonal directions (front-back, side to side, and up-down), at four levels of peak acceleration (0.1g, 0.5g, 1.0g, and 3.0g), each repeated five times resulting in a total of 60 tests per unit, for a total of 1140 records. Data from each accelerometer was recorded at a sampling frequency of 100Hz. Peak accelerations recorded by the devices, Catapult PlayerLoad™, and calculated player load (using Catapult’s Cartesian formula) were used for the analysis. The devices demonstrated excellent intradevice reliability and mixed interdevice reliability. Differences were found between devices for mean peak accelerations and PlayerLoad™ for each direction and level of acceleration. Interdevice effect sizes ranged from a mean of 0.54 (95% CI: 0.34–0.74) (small) to 1.20 (95% CI: 1.08–1.30) (large) and ICCs ranged from 0.77 (95% CI: 0.62–0.89) (very large) to 1.0 (95% CI: 0.99–1.0) (nearly perfect) depending upon the magnitude and direction of the applied motion. When compared to the player load determined using the Cartesian formula, the Catapult reported PlayerLoad™ was consistently lower by approximately 15%. These results emphasize the need for industry wide standards in reporting validity, reliability and the magnitude of measurement errors. It is recommended that device reliability and accuracy are periodically quantified.

Evaluation of sampling frequency, window size and sensor position for classification of sheep behaviour

Automated behavioural classification and identification through sensors has the potential to improve health and welfare of the animals. Position of a sensor, sampling frequency and window size of segmented signal data has a major impact on classification accuracy in activity recognition and energy needs for the sensor, yet, there are no studies in precision livestock farming that have evaluated the effect of all these factors simultaneously. The aim of this study was to evaluate the effects of position (ear and collar), sampling frequency (8, 16 and 32 Hz) of a triaxial accelerometer and gyroscope sensor and window size (3, 5 and 7 s) on the classification of important behaviours in sheep such as lying, standing and walking. Behaviours were classified using a random forest approach with 44 feature characteristics. The best performance for walking, standing and lying classification in sheep (accuracy 95%, F-score 91%-97%) was obtained using combination of 32 Hz, 7 s and 32 Hz, 5 s for both ear and collar sensors, although, results obtained with 16 Hz and 7 s window were comparable with accuracy of 91%-93% and F-score 88%-95%. Energy efficiency was best at a 7 s window. This suggests that sampling at 16 Hz with 7 s window will offer benefits in a real-time behavioural monitoring system for sheep due to reduced energy needs.

The relationship between movement speed and duration during soccer matches

The relationship between the time duration of movement (t(dur)) and related maximum possible power output has been studied and modeled under many conditions. Inspired by the so-called power profiles known for discontinuous endurance sports like cycling, and the critical power concept of Monod and Scherrer, the aim of this study was to evaluate the numerical characteristics of the function between maximum horizontal movement velocity (HSpeed) and t(dur) in soccer. To evaluate this relationship, GPS data from 38 healthy soccer players and 82 game participations (≥30 min active playtime) were used to select maximum HSpeed for 21 distinct t(dur) values (between 0.3 s and 2,700 s) based on moving medians with an incremental t(dur) window-size. As a result, the relationship between HSpeed and Log(t(dur)) appeared reproducibly as a sigmoidal decay function, and could be fitted to a five-parameter equation with upper and lower asymptotes, and an inflection point, power and decrease rate. Thus, the first three parameters described individual characteristics if evaluated using mixed-model analysis. This study shows for the first time the general numerical relationship between t(dur) and HSpeed in soccer games. In contrast to former descriptions that have evaluated speed against power, HSpeed against t(dur) always yields a sigmoidal shape with a new upper asymptote. The evaluated curve fit potentially describes the maximum moving speed of individual players during the game, and allows for concise interpretations of the functional state of team sports athletes.

When Is a Sprint a Sprint? A Review of the Analysis of Team-Sport Athlete Activity Profile

The external load of a team-sport athlete can be measured by tracking technologies, including global positioning systems (GPS), local positioning systems (LPS), and vision-based systems. These technologies allow for the calculation of displacement, velocity and acceleration during a match or training session. The accurate quantification of these variables is critical so that meaningful changes in team-sport athlete external load can be detected. High-velocity running, including sprinting, may be important for specific team-sport match activities, including evading an opponent or creating a shot on goal. Maximal accelerations are energetically demanding and frequently occur from a low velocity during team-sport matches. Despite extensive research, conjecture exists regarding the thresholds by which to classify the high velocity and acceleration activity of a team-sport athlete. There is currently no consensus on the definition of a sprint or acceleration effort, even within a single sport. The aim of this narrative review was to examine the varying velocity and acceleration thresholds reported in athlete activity profiling. The purposes of this review were therefore to (1) identify the various thresholds used to classify high-velocity or -intensity running plus accelerations; (2) examine the impact of individualized thresholds on reported team-sport activity profile; (3) evaluate the use of thresholds for court-based team-sports and; (4) discuss potential areas for future research. The presentation of velocity thresholds as a single value, with equivocal qualitative descriptors, is confusing when data lies between two thresholds. In Australian football, sprint efforts have been defined as activity >4.00 or >4.17 m·s⁻¹. Acceleration thresholds differ across the literature, with >1.11, 2.78, 3.00, and 4.00 m·s⁻² utilized across a number of sports. It is difficult to compare literature on field-based sports due to inconsistencies in velocity and acceleration thresholds, even within a single sport. Velocity and acceleration thresholds have been determined from physical capacity tests. Limited research exists on the classification of velocity and acceleration data by female team-sport athletes. Alternatively, data mining techniques may be used to report team-sport athlete external load, without the requirement of arbitrary or physiologically defined thresholds.

Attacking 22 entries in rugby union: running demands and differences between successful and unsuccessful entries

Global Positioning System (GPS) technology is commonly utilized in team sports, including rugby union. It has been used to describe the average running demands of rugby union. This has afforded an enhanced understanding of the physical fitness requirements for players. However, research in team sports has suggested that training players relative to average demands may underprepare them for certain scenarios within the game. To date, no research has investigated the running demands of attacking 22 entries in rugby union. Additionally, no research has been undertaken to determine whether differences exist in the running intensity of successful and unsuccessful attacking 22 entries in rugby union. The first aim of this study was to describe the running intensity of attacking 22 entries. The second aim of this study was to investigate whether differences exist in the running intensity of successful and unsuccessful attacking 22 entries. Running intensity was measured using meters per minute (m min^-1 ) for (a) total distance, (b) running distance, (c) high-speed running distance, and (d) very high-speed running distance. This study provides normative data for the running intensity of attacking 22 entries in rugby union. Forwards achieved greater high-speed running intensity in successful (3.6 m min^-1 ) compared to unsuccessful (1.8 m min^-1 ) attacking 22 entries. Forwards should try and achieve greater high-speed running intensity in attacking 22 entries to increase the likelihood of successful outcomes during this period of gameplay.