Machine Learning Based-prediction of Health Application Effectiveness on Google Play Store

Abstract

This study aims to evaluate the effectiveness of health applications on the Google Play Store by analyzing app metadata using machine learning classification models. It investigates whether application features—such as classification, app category, update status, and version—are associated with higher user ratings. A total of 305 health-related applications were selected from the Google Play Store using keyword filters for “Health & Fitness” and “Medical.” Key metadata were extracted and pre-processed, including Classification (AI vs. Non-AI), Category, Reviews, Developer Type, Version, Release Year, and Recent Update. To address class imbalance, the SMOTE technique was applied, and three machine learning models—Naïve Bayes and K-Nearest Neighbors (KNN)—were used to predict user ratings. The KNN model achieved the most balanced performance with 75.89% accuracy, 82.22% precision, and an AUC of 0.849. Future research should consider larger and more diverse datasets and explore additional features (e.g., user sentiment from reviews, app permissions) to further improve model performance.