Evaluating and implementing machine learning models for personalised mobile health app recommendations.

Area of Science:

• Digital Health
• Health Informatics
• Machine Learning in Healthcare

Background:

• The proliferation of mobile health (mHealth) applications necessitates effective evaluation and recommendation systems.
• A growing mHealth market underscores the need for tools to guide users to relevant health applications.

Purpose of the Study:

• To design and evaluate a recommendation system for mHealth applications.
• To leverage app attributes, particularly descriptions, for contextual user suggestions.
• To categorize health applications for improved user experience.

Main Methods:

• A dataset of health app attributes, including descriptions, ratings, and reviews, was curated.
• Data pre-processing involved one-hot encoding, standardization, and feature engineering, including a novel 'Rating_Reviews' feature.
• Machine learning and deep learning models, including Random Forest and BERT, were evaluated using 'Category' (e.g., 'Weight loss', 'Medical') as the target variable.

Main Results:

• The BERT model, utilizing transfer learning, demonstrated high efficiency, achieving approximately 90% accuracy after hyperparameter tuning.
• Feature engineering, including the 'Rating_Reviews' metric, contributed to model performance.
• The 'Category' variable effectively distinguished different health contexts within the mHealth landscape.

Conclusions:

• Transfer learning, particularly with the BERT model, is highly effective for mHealth application recommendation.
• A robust recommendation system can be built using app descriptions and user-centric features.
• The developed system, employing cosine similarity, accurately ranks apps based on user query relevance.