insamhlaithe/Machine-Learning-for-ECG-Based-Human-Func-tional-State-Assessment

Code files for the article "Machine Learning Framework for ECG-Based Human Func-tional State Assessment Using Interpretable Feature Analysis And Unsupervised Clustering"

Abstract This repository contains the code and materials developed by the authors for a hybrid machine learning pipeline designed to assess functional states from ECG data. The pipeline integrates feature-based supervised classification with interpretable analyses and unsupervised clustering techniques. The underlying dataset, gathered under field conditions from military personnel, injured service members, rehabilitation cohorts, and civilian controls, was thoroughly preprocessed — including imputation of missing values, outlier handling, correlation filtering, and normalization — resulting in 17 physiologically and psychophysiologically relevant features. Using HDBSCAN clustering, two main macro-groups emerged: one representing frontline and training personnel, and another including hospital and rehabilitation participants; civilian controls displayed high variability and were largely excluded from downstream analyses. Gradient boosting classifiers achieved strong predictive performance (≈0.97–0.98 accuracy), confirming the clear separability of the clusters. SHAP analysis highlighted S-wave amplitude (lead I), J40 amplitude, and the composite MYOCARDIUM index as primary discriminators, while psycho-emotional state, HRV total power, and repolarization measures provided additional insight. These results suggest that the identified features reflect meaningful physiological differences between active-duty and recovery states. Overall, this work demonstrates that combining interpretable machine learning with nonlinear embedding and density-based clustering allows both accurate classification and physiologically interpretable insights. This repository serves as a resource for extending ECG analysis toward monitoring functional readiness, stress, and adaptation, with applications in military and occupational health contexts.

No dataset available.