Comparative Study of Machine Learning Methods for the Early Prediction of Adherence to Medication
摘要
Adherence to medication is a critical aspect of healthcare with a significant impact on patient outcomes. This has led to the elaboration of several studies over the years to understand adherence better, evolving to the point of applying Machine Learning techniques, whether to study the relationship of different factors with adherence or to make predictions of levels of adherence. However, due to the diversity of techniques, evaluation metrics, and adherence measures utilized, no conclusions have been drawn as to which algorithms are best suited to address prediction problems in this domain. This paper aims to apply the three most widely used algorithms in the literature to a database obtained from a primary care center. The study evaluates the performance of these algorithms with and without applying a feature selection method and with three different adherence measures, using four different evaluation metrics. The findings reveal that Logistic Regression consistently achieves superior performance, with evaluation metrics exceeding 0.90 in the majority of the scenarios, thereby underscoring its efficacy in predicting adherence to medication.