Development and Evaluation of a Machine Learning Framework for Predicting Patient Adherence to Mobile Health Applications in Chronic Disease Management

Abstract

Chronic diseases constitute one of the foremost public health challenges of the twenty-first century, and mobile health (mHealth) applications have emerged as promising adjuncts to conventional care. However, inconsistent and unsustained patient engagement with these tools substantially diminishes their clinical value. Predicting which patients are likely to sustain engagement with mHealth applications remains an underexplored area of health informatics research, particularly in low- and middle-income country (LMIC) settings. This study sought to develop and evaluate a machine learning framework capable of predicting patient adherence to mHealth applications for chronic disease management. A cross-sectional survey design was employed, and data were collected from 487 patients recruited from tertiary health facilities across three sub-Saharan African countries. Nine predictor variables were examined: age, gender, education, health literacy, trust, privacy concern, disease type, duration of illness, and social influence. The outcome variable, mHealth adherence, was operationalized into three categories High Usage, Low Usage, and Sustained Engagement. Five algorithms were developed and compared: Logistic Regression, Random Forest, Support Vector Machine, Artificial Neural Network, and XGBoost. Data preprocessing involved imputation of missing values, outlier detection, feature encoding, normalization, and feature engineering. All models were trained on an 80/20 train-test split with stratified ten-fold cross-validation and grid-search hyperparameter optimization. XGBoost achieved the highest performance (Accuracy = 93.8%, F1 Score = 0.937, AUC-ROC = 0.978), followed by Random Forest (Accuracy = 91.2%), and Artificial Neural Network (Accuracy = 88.6%). Health literacy, trust, and social influence were the most influential predictors. These findings underscore the viability of ensemble-based machine learning in supporting clinical decision-making for chronic disease mHealth adherence and offer actionable insights for application designers, healthcare providers, and policymakers.