Efficient Leave-One-Out Cross-Validation-based Regularized Extreme Learning Machine

Predicting branch retinal vein occlusion development using multimodal deep learning and pre-onset fundus hemisection images

Branch retinal vein occlusion (BRVO) is a leading cause of visual impairment in working-age individuals, though predicting its occurrence from retinal vascular features alone remains challenging. We developed a deep learning model to predict BRVO based on pre-onset, metadata-matched fundus hemisection images. This retrospective cohort study included patients diagnosed with unilateral BRVO from two Korean tertiary centers (2005-2023), using hemisection fundus images from 27 BRVO-affected eyes paired with 81 unaffected hemisections (27 counter and 54 contralateral) for training. A U-net model segmented retinal optic discs and blood vessels (BVs), dividing them into upper and lower halves labeled for BRVO occurrence. Both unimodal models (using either fundus or BV images) and a BV-enhanced multimodal model were constructed to predict future BRVO. The multimodal model outperformed the unimodal models achieving an area under the receiver operating characteristic curve of 0.76 (95% confidence interval [CI], 0.66-0.83) and accuracy of 68.5% (95% CI 58.9-77.1%), with predictions focusing on arteriovenous crossing regions in the retinal vascular arcade. These findings demonstrate the potential of the BV-enhanced multimodal approach for BRVO prediction and highlight the need for larger, multicenter datasets to improve its clinical utility and predictive accuracy.

A Survey on Scenario Theory, Complexity, and Compression-Based Learning and Generalization

This work investigates formal generalization error bounds that apply to support vector machines (SVMs) in realizable and agnostic learning problems. We focus on recently observed parallels between probably approximately correct (PAC)-learning bounds, such as compression and complexity-based bounds, and novel error guarantees derived within scenario theory. Scenario theory provides nonasymptotic and distributional-free error bounds for models trained by solving data-driven decision-making problems. Relevant theorems and assumptions are reviewed and discussed. We propose a numerical comparison of the tightness and effectiveness of theoretical error bounds for support vector classifiers trained on several randomized experiments from 13 real-life problems. This analysis allows for a fair comparison of different approaches from both conceptual and experimental standpoints. Based on the numerical results, we argue that the error guarantees derived from scenario theory are often tighter for realizable problems and always yield informative results, i.e., probability bounds tighter than a vacuous [0, 1] interval. This work promotes scenario theory as an alternative tool for model selection, structural-risk minimization, and generalization error analysis of SVMs. In this way, we hope to bring the communities of scenario and statistical learning theory closer, so that they can benefit from each other's insights.

Developing a Prediction Model of Demolition-Waste Generation-Rate via Principal Component Analysis

Construction and demolition waste accounts for a sizable proportion of global waste and is harmful to the environment. Its management is therefore a key challenge in the construction industry. Many researchers have utilized waste generation data for waste management, and more accurate and efficient waste management plans have recently been prepared using artificial intelligence models. Here, we developed a hybrid model to forecast the demolition-waste-generation rate in redevelopment areas in South Korea by combining principal component analysis (PCA) with decision tree, k-nearest neighbors, and linear regression algorithms. Without PCA, the decision tree model exhibited the highest predictive performance (R2 = 0.872) and the k-nearest neighbors (Chebyshev distance) model exhibited the lowest (R2 = 0.627). The hybrid PCA-k-nearest neighbors (Euclidean uniform) model exhibited significantly better predictive performance (R2 = 0.897) than the non-hybrid k-nearest neighbors (Euclidean uniform) model (R2 = 0.664) and the decision tree model. The mean of the observed values, k-nearest neighbors (Euclidean uniform) and PCA-k-nearest neighbors (Euclidean uniform) models were 987.06 (kg·m-2), 993.54 (kg·m-2) and 991.80 (kg·m-2), respectively. Based on these findings, we propose the k-nearest neighbors (Euclidean uniform) model using PCA as a machine-learning model for demolition-waste-generation rate predictions.

Diagnosis of T-cell-mediated kidney rejection by biopsy-based proteomic biomarkers and machine learning

Background

Biopsy-based diagnosis is essential for maintaining kidney allograft longevity by ensuring prompt treatment for graft complications. Although histologic assessment remains the gold standard, it carries significant limitations such as subjective interpretation, suboptimal reproducibility, and imprecise quantitation of disease burden. It is hoped that molecular diagnostics could enhance the efficiency, accuracy, and reproducibility of traditional histologic methods.

Methods

Quantitative label-free mass spectrometry analysis was performed on a set of formalin-fixed, paraffin-embedded (FFPE) biopsies from kidney transplant patients, including five samples each with diagnosis of T-cell-mediated rejection (TCMR), polyomavirus BK nephropathy (BKPyVN), and stable (STA) kidney function control tissue. Using the differential protein expression result as a classifier, three different machine learning algorithms were tested to build a molecular diagnostic model for TCMR.

Results

The label-free proteomics method yielded 800-1350 proteins that could be quantified with high confidence per sample by single-shot measurements. Among these candidate proteins, 329 and 467 proteins were defined as differentially expressed proteins (DEPs) for TCMR in comparison with STA and BKPyVN, respectively. Comparing the FFPE quantitative proteomics data set obtained in this study using label-free method with a data set we previously reported using isobaric labeling technology, a classifier pool comprised of features from DEPs commonly quantified in both data sets, was generated for TCMR prediction. Leave-one-out cross-validation result demonstrated that the random forest (RF)-based model achieved the best predictive power. In a follow-up blind test using an independent sample set, the RF-based model yields 80% accuracy for TCMR and 100% for STA. When applying the established RF-based model to two public transcriptome datasets, 78.1%-82.9% sensitivity and 58.7%-64.4% specificity was achieved respectively.

Conclusions

This proof-of-principle study demonstrates the clinical feasibility of proteomics profiling for FFPE biopsies using an accurate, efficient, and cost-effective platform integrated of quantitative label-free mass spectrometry analysis with a machine learning-based diagnostic model. It costs less than 10 dollars per test.

Development of Machine Learning Model for Prediction of Demolition Waste Generation Rate of Buildings in Redevelopment Areas

Owing to a rapid increase in waste, waste management has become essential, for which waste generation (WG) information has been effectively utilized. Various studies have recently focused on the development of reliable predictive models by applying artificial intelligence to the construction and prediction of WG information. In this study, research was conducted on the development of machine learning (ML) models for predicting the demolition waste generation rate (DWGR) of buildings in redevelopment areas in South Korea. Various ML algorithms (i.e., artificial neural network (ANN), K-nearest neighbors (KNN), linear regression (LR), random forest (RF), and support vector machine (SVM)) were applied to the development of an optimal predictive model, and the main hyper parameters (HPs) for each algorithm were optimized. The results suggest that ANN-ReLu (coefficient of determination (R²) 0.900, the ratio of percent deviation (RPD) 3.16), SVM-polynomial (R² 0.889, RPD 3.00), and ANN-logistic (R² 0.883, RPD 2.92) are the best ML models for predicting the DWGR. They showed average errors of 7.3%, 7.4%, and 7.5%, respectively, compared to the average observed values, confirming the accurate predictive performance, and in the uncertainty analysis, the d-factor of the models appeared less than 1, showing that the presented models are reliable. Through a comparison with ML algorithms and HPs applied in previous related studies, the results herein also showed that the selection of various ML algorithms and HPs is important in developing optimal ML models for WG management.

Adult Skeletal Age-at-Death Estimation through Deep Random Neural Networks: A New Method and Its Computational Analysis

Age-at-death assessment is a crucial step in the identification process of skeletal human remains. Nonetheless, in adult individuals this task is particularly difficult to achieve with reasonable accuracy due to high variability in the senescence processes. To improve the accuracy of age-at-estimation, in this work we propose a new method based on a multifactorial macroscopic analysis and deep random neural network models. A sample of 500 identified skeletons was used to establish a reference dataset (age-at-death: 19–101 years old, 250 males and 250 females). A total of 64 skeletal traits are covered in the proposed macroscopic technique. Age-at-death estimation is tackled from a function approximation perspective and a regression approach is used to infer both point and prediction interval estimates. Based on cross-validation and computational experiments, our results demonstrate that age estimation from skeletal remains can be accurately (~6 years mean absolute error) inferred across the entire adult age span and informative estimates and prediction intervals can be obtained for the elderly population. A novel software tool, DRNNAGE, was made available to the community.

Artificial Intelligence Applied to Battery Research: Hype or Reality?

This is a critical review of artificial intelligence/machine learning (AI/ML) methods applied to battery research. It aims at providing a comprehensive, authoritative, and critical, yet easily understandable, review of general interest to the battery community. It addresses the concepts, approaches, tools, outcomes, and challenges of using AI/ML as an accelerator for the design and optimization of the next generation of batteries—a current hot topic. It intends to create both accessibility of these tools to the chemistry and electrochemical energy sciences communities and completeness in terms of the different battery R&D aspects covered.

Comparison of Random Forest and Gradient Boosting Machine Models for Predicting Demolition Waste Based on Small Datasets and Categorical Variables

Construction and demolition waste (DW) generation information has been recognized as a tool for providing useful information for waste management. Recently, numerous researchers have actively utilized artificial intelligence technology to establish accurate waste generation information. This study investigated the development of machine learning predictive models that can achieve predictive performance on small datasets composed of categorical variables. To this end, the random forest (RF) and gradient boosting machine (GBM) algorithms were adopted. To develop the models, 690 building datasets were established using data preprocessing and standardization. Hyperparameter tuning was performed to develop the RF and GBM models. The model performances were evaluated using the leave-one-out cross-validation technique. The study demonstrated that, for small datasets comprising mainly categorical variables, the bagging technique (RF) predictions were more stable and accurate than those of the boosting technique (GBM). However, GBM models demonstrated excellent predictive performance in some DW predictive models. Furthermore, the RF and GBM predictive models demonstrated significantly differing performance across different types of DW. Certain RF and GBM models demonstrated relatively low predictive performance. However, the remaining predictive models all demonstrated excellent predictive performance at R2 values > 0.6, and R values > 0.8. Such differences are mainly because of the characteristics of features applied to model development; we expect the application of additional features to improve the performance of the predictive models. The 11 DW predictive models developed in this study will be useful for establishing detailed DW management strategies.

Development of a Prediction Model for Demolition Waste Generation Using a Random Forest Algorithm Based on Small DataSets

Recently, artificial intelligence (AI) technologies have been employed to predict construction and demolition (C&D) waste generation. However, most studies have used machine learning models with continuous data input variables, applying algorithms, such as artificial neural networks, adaptive neuro-fuzzy inference systems, support vector machines, linear regression analysis, decision trees, and genetic algorithms. Therefore, machine learning algorithms may not perform as well when applied to categorical data. This article uses machine learning algorithms to predict C&D waste generation from a dataset, as a way to improve the accuracy of waste management in C&D facilities. These datasets include categorical (e.g., region, building structure, building use, wall material, and roofing material), and continuous data (particularly, gloss floor area), and a random forest (RF) algorithm was used. Results indicate that RF is an adequate machine learning algorithm for a small dataset consisting of categorical data, and even with a small dataset, an adequate prediction model can be developed. Despite the small dataset, the predictive performance according to the demolition waste (DW) type was R (Pearson's correlation coefficient) = 0.691-0.871, R2 (coefficient of determination) = 0.554-0.800, showing stable prediction performance. High prediction performance was observed using three (for mortar), five (for other DW types), or six (for concrete) input variables. This study is significant because the proposed RF model can predict DW generation using a small amount of data. Additionally, it demonstrates the possibility of applying AI to multi-purpose DW management.

A fast conformal predictive system with regularized extreme learning machine

A conformal predictive system(CPS) is based on the learning framework of conformal prediction, which outputs cumulative distribution functions(CDFs) for labels in regression problems. The CDFs output by a CPS provide useful information for users, as they not only provide probability for the events related to the test labels, but also can be transformed to prediction intervals with the corresponding quantiles. Moreover, CPSs have the property of validity since the distributions and intervals they output have statistical compatibility with the realizations. This property is very useful for many risk-sensitive applications such as financial time series forecast and weather forecast. However, as based on conformal predictors, CPSs inherit the computational issue. To build a fast CPS, in this paper, we propose a CPS with regularized extreme learning machine as the underlying algorithm. To be specific, we combine the leave-one-out cross-conformal predictive system(Leave-One-Out CCPS), a variant of the original CPS, with regularized extreme learning machine(RELM), which is named as LOO-CCPS-RELM. We analyse the computational complexity of it and prove its asymptotic validity based on some regularity assumptions. We also prove that the error rate of the prediction interval output by LOO-CCPS-RELM is under control in the asymptotic setting. Experiments with 20 public data sets were conducted to test LOO-CCPS-RELM and the results showed that LOO-CCPS-RELM is empirically valid and compared favourably with the other CPSs.

Development of artificial neural network model for prediction of post-streptococcus mutans in dental caries

BACKGROUND AND OBJECTIVE

Streptococcus mutans is the primary initiator and most common organism associated with dental caries. Prediction of post-Streptococcus mutans favours in the selection of appropriate caries excavation method which eventually results in meliorate caries-free cavity preparation for restoration. The objective of this study is to predict the post-Streptococcus mutans prior to dental caries excavation based on pre- Streptococcus mutans using iOS App developed on Artificial Neural Network (ANN) model.

METHODS

For the current research work, children with occlusal dentinal caries lesion were chosen, 45 primary molar teeth cases were studied. Caries excavation was done with carbide bur, polymer bur and spoon excavator. The colony forming units for pre and post-Streptococcus mutans were recorded, data emanating from clinical trials was employed to develop the ANN models. ANN models were trained, validated and tested with the registered clinical data using different ANN architectures.

RESULTS

Feedforward backpropagation ANN model with an architecture of 4-5-1, predicts post-Streptococcus mutans with an efficiency of 0.99033, mean squared error and mean absolute percentage error for testing cases were 0.2341 and 4.967 respectively.

CONCLUSIONS

Caries excavation methods and pre-Streptococcus mutans are feed as inputs, while post-Streptococcus mutans as targets to develop ANN model. Based on the developed ANN model, an ingenious iOS App was developed, the global clinician may utilize the App to meticulously predict post-Streptococcus mutans on iPhone based on pre-Streptococcus mutans, which in turn aids in decision making for the selection of caries excavation method. This study manifests the potential application of iOS App with built-in ANN model in efficiently predicting the post-Streptococcus mutans. Also, the study extends scope for applications of iOS App with built-in ANN models in clinical medicine.

Laser-Driven Calorimetry and Chemometric Quantification of Standard Reference Material Diesel/Biodiesel Fuel Blends

Requirements for blends of drop-in petroleum/bio-derived fuels with specific thermophysical and thermochemical properties highlights the need for chemometric models that can predict these properties. Multivariate calibration methods were evaluated using the measured thermograms (i.e., change in temperature with time) of 11 diesel/biodiesel fuel blends (including four repeated runs for each fuel blend). Two National Institute of Standards and Technology Standard Reference Material® (SRM®) pure fuels were blended by serial dilution to produce fuels having diesel/biodiesel volumetric fractions between (0 to 100) %. The fuels were evaluated for the prepared fuel-blend volume fraction and total specific energy release (heating value), using a laser-driven calorimetry technique, termed 'laser-driven thermal reactor'. The experimental apparatus consists of a copper sphere-shaped reactor (mounted at the center of a stainless-steel chamber) that is heated by a high-power continuous wave Nd:YAG laser. Prior to heating by the laser, liquid sample is injected onto a copper pan substrate that rests near the center of the reactor and is in contact with a fine-wire thermocouple. A second thermocouple is in contact with the sphere-reactor inner surface. The thermograms are then used to evaluate for the thermochemical characteristic of interest. Partial least squares (PLS) and support vector machine (SVM) models were constructed and evaluated for SRM-fuel-blend quantification, and determination of prepared fuel-blend volume fraction and heating value. Quantification of the fuel-blend thermograms by the SVM method was found to better correlate with the experimental results than PLS. The combination of laser-driven calorimetry and multivariate calibration methods has demonstrated the potential application of using thermograms for fuels quantification and analysis of fuel-blend properties.

Quantitative Structure-activity Relationship Analysis for Predicting Lipophilicity of Aniline Derivatives (Including some Pharmaceutical Compounds)

BACKGROUND

In this study, we used a hierarchical approach to develop quantitative structureactivity relationship (QSAR) models for modeling lipophilicity of a set of 81 aniline derivatives containing some pharmaceutical compounds.

OBJECTIVE

The multiple linear regression (MLR), principal component regression (PCR) and partial least square regression (PLSR) methods were utilized to construct QSAR models.

MATERIALS AND METHODS

Quantum mechanical calculations at the density functional theory level and 6- 311++G** basis set were carried out to obtain the optimized geometry and then, the comprehensive set of molecular descriptors was computed by using the Dragon software. Genetic algorithm (GA) was applied to select suitable descriptors which have the most correlation with lipophilicity of the studied compounds.

RESULTS

It was identified that such descriptors as Barysz matrix (SEigZ), hydrophilicity factor (Hy), Moriguchi octanol-water partition coefficient (MLOGP), electrophilicity (ω/eV) van der Waals volume (vWV) and lethal concentration (LC50/molkg-1) are the best descriptors for QSAR modeling. The high correlation coefficients and the low prediction errors for MLR, PCR and PLSR methods confirmed good predictability of the three models.

CONCLUSION

In present study, the high correlation between experimental and predicted logP values of aniline derivatives indicated the validation and the good quality of the resulting three regression methods, but MLR regression procedure was a little better than the PCR and PLSR methods. It was concluded that the studied aniline derivatives are not hydrophilic compounds and this means these compounds hardly dissolve in water or an aqueous solvent.