Support Vector Machine Classifiers Show High Generalizability in Automatic Fall Detection in Older Adults

Development and validation of machine learning models for nonalcoholic fatty liver disease

BACKGROUND

Nonalcoholic fatty liver disease (NAFLD) had become the most prevalent liver disease worldwide. Early diagnosis could effectively reduce NAFLD-related morbidity and mortality. This study aimed to combine the risk factors to develop and validate a novel model for predicting NAFLD.

METHODS

We enrolled 578 participants completing abdominal ultrasound into the training set. The least absolute shrinkage and selection operator (LASSO) regression combined with random forest (RF) was conducted to screen significant predictors for NAFLD risk. Five machine learning models including logistic regression (LR), RF, extreme gradient boosting (XGBoost), gradient boosting machine (GBM), and support vector machine (SVM) were developed. To further improve model performance, we conducted hyperparameter tuning with train function in Python package 'sklearn'. We included 131 participants completing magnetic resonance imaging into the testing set for external validation.

RESULTS

There were 329 participants with NAFLD and 249 without in the training set, while 96 with NAFLD and 35 without were in the testing set. Visceral adiposity index, abdominal circumference, body mass index, alanine aminotransferase (ALT), ALT/AST (aspartate aminotransferase), age, high-density lipoprotein cholesterol (HDL-C) and elevated triglyceride (TG) were important predictors for NAFLD risk. The area under curve (AUC) of LR, RF, XGBoost, GBM, SVM were 0.915 [95% confidence interval (CI): 0.886-0.937], 0.907 (95% CI: 0.856-0.938), 0.928 (95% CI: 0.873-0.944), 0.924 (95% CI: 0.875-0.939), and 0.900 (95% CI: 0.883-0.913), respectively. XGBoost model presented the best predictive performance, and its AUC was enhanced to 0.938 (95% CI: 0.870-0.950) with further parameter tuning.

CONCLUSIONS

This study developed and validated five novel machine learning models for NAFLD prediction, among which XGBoost presented the best performance and was considered a reliable reference for early identification of high-risk patients with NAFLD in clinical practice.

CNN-Based Self-Attention Weight Extraction for Fall Event Prediction Using Balance Test Score

Injury, hospitalization, and even death are common consequences of falling for elderly people. Therefore, early and robust identification of people at risk of recurrent falling is crucial from a preventive point of view. This study aims to evaluate the effectiveness of an interpretable semi-supervised approach in identifying individuals at risk of falls by using the data provided by ankle-mounted IMU sensors. Our method benefits from the cause-effect link between a fall event and balance ability to pinpoint the moments with the highest fall probability. This framework also has the advantage of training on unlabeled data, and one can exploit its interpretation capacities to detect the target while only using patient metadata, especially those in relation to balance characteristics. This study shows that a visual-based self-attention model is able to infer the relationship between a fall event and loss of balance by attributing high values of weight to moments where the vertical acceleration component of the IMU sensors exceeds 5 m/s² during an especially short period. This semi-supervised approach uses interpretable features to highlight the moments of the recording that may explain the score of balance, thus revealing the moments with the highest risk of falling. Our model allows for the detection of 71% of the possible falling risk events in a window of 1 s (500 ms before and after the target) when compared with threshold-based approaches. This type of framework plays a paramount role in reducing the costs of annotation in the case of fall prevention when using wearable devices. Overall, this adaptive tool can provide valuable data to healthcare professionals, and it can assist them in enhancing fall prevention efforts on a larger scale with lower costs.

Patch-Transformer Network: A Wearable-Sensor-Based Fall Detection Method

Falls can easily cause major harm to the health of the elderly, and timely detection can avoid further injuries. To detect the occurrence of falls in time, we propose a new method called Patch-Transformer Network (PTN) wearable-sensor-based fall detection algorithm. The neural network includes a convolution layer, a Transformer encoding layer, and a linear classification layer. The convolution layer is used to extract local features and project them into feature matrices. After adding positional coding information, the global features of falls are learned through the multi-head self-attention mechanism in the Transformer encoding layer. Global average pooling (GAP) is used to strengthen the correlation between features and categories. The final classification results are provided by the linear layer. The accuracy of the model obtained on the public available datasets SisFall and UnMib SHAR is 99.86% and 99.14%, respectively. The network model has fewer parameters and lower complexity, with detection times of 0.004 s and 0.001 s on the two datasets. Therefore, our proposed method can timely and accurately detect the occurrence of falls, which is important for protecting the lives of the elderly.

Image Geo-Site Estimation Using Convolutional Auto-Encoder and Multi-Label Support Vector Machine

The estimation of an image geo-site solely based on its contents is a promising task. Compelling image labelling relies heavily on contextual information, which is not as simple as recognizing a single object in an image. An Auto-Encode-based support vector machine approach is proposed in this work to estimate the image geo-site to address the issue of misclassifying the estimations. The proposed method for geo-site estimation is conducted using a dataset consisting of 125 classes of various images captured within 125 countries. The proposed work uses a convolutional Auto-Encode for training and dimensionality reduction. After that, the acquired preprocessed input dataset is further processed by a multi-label support vector machine. The performance assessment of the proposed approach has been accomplished using accuracy, sensitivity, specificity, and F1-score as evaluation parameters. Eventually, the proposed approach for image geo-site estimation presented in this article outperforms Auto-Encode-based K-Nearest Neighbor and Auto-Encode-Random Forest methods.

Prediction of HIV-1 protease cleavage site from octapeptide sequence information using selected classifiers and hybrid descriptors

Background

In most parts of the world, especially in underdeveloped countries, acquired immunodeficiency syndrome (AIDS) still remains a major cause of death, disability, and unfavorable economic outcomes. This has necessitated intensive research to develop effective therapeutic agents for the treatment of human immunodeficiency virus (HIV) infection, which is responsible for AIDS. Peptide cleavage by HIV-1 protease is an essential step in the replication of HIV-1. Thus, correct and timely prediction of the cleavage site of HIV-1 protease can significantly speed up and optimize the drug discovery process of novel HIV-1 protease inhibitors. In this work, we built and compared the performance of selected machine learning models for the prediction of HIV-1 protease cleavage site utilizing a hybrid of octapeptide sequence information comprising bond composition, amino acid binary profile (AABP), and physicochemical properties as numerical descriptors serving as input variables for some selected machine learning algorithms. Our work differs from antecedent studies exploring the same subject in the combination of octapeptide descriptors and method used. Instead of using various subsets of the dataset for training and testing the models, we combined the dataset, applied a 3-way data split, and then used a "stratified" 10-fold cross-validation technique alongside the testing set to evaluate the models.

Results

Among the 8 models evaluated in the “stratified” 10-fold CV experiment, logistic regression, multi-layer perceptron classifier, linear discriminant analysis, gradient boosting classifier, Naive Bayes classifier, and decision tree classifier with AUC, F-score, and B. Acc. scores in the ranges of 0.91–0.96, 0.81–0.88, and 80.1–86.4%, respectively, have the closest predictive performance to the state-of-the-art model (AUC 0.96, F-score 0.80 and B. Acc. ~ 80.0%). Whereas, the perceptron classifier and the K-nearest neighbors had statistically lower performance (AUC 0.77–0.82, F-score 0.53–0.69, and B. Acc. 60.0–68.5%) at p < 0.05. On the other hand, logistic regression, and multi-layer perceptron classifier (AUC of 0.97, F-score > 0.89, and B. Acc. > 90.0%) had the best performance on further evaluation on the testing set, though linear discriminant analysis, gradient boosting classifier, and Naive Bayes classifier equally performed well (AUC > 0.94, F-score > 0.87, and B. Acc. > 86.0%).

Conclusions

Logistic regression and multi-layer perceptron classifiers have comparable predictive performances to the state-of-the-art model when octapeptide sequence descriptors consisting of AABP, bond composition and standard physicochemical properties are used as input variables. In our future work, we hope to develop a standalone software for HIV-1 protease cleavage site prediction utilizing the linear regression algorithm and the aforementioned octapeptide sequence descriptors.