Predicting the Adherence to Physical Activity using Ensemble Machine Learning

Several studies have reported that physical exercise and activity play a crucial role in an individual's health and well-being. This study aims to effectively identify and reduce temporary and permanent exercise relapses of everyday physical activity using ensemble classifiers. The dataset used contains the daily step count of several subjects. These are aggregated into a weekly step count. The first week's step count is used as the initial or run-in period step count, which is compared with the step counts of the following weeks to determine the Discontinuation Prediction Score’ (DiPS). The ensemble model learns the data and predicts the possibility of exercise relapse for the subsequent weeks. The prediction accuracy of the ensemble model is compared with other classification models such as logistic regression and support vector machines. It is observed that the ensemble classifier has higher accuracy, specificity and sensitivity than that of the other models. Thus, we conclude that ensemble classifiers are better at predicting breaks in an individual's adherence to their physical activity regimen.