ML-Driven Health Prediction using Smartphone Screen Time

Abstract

In recent years, research has shown a direct link between how much time we use our smartphones and our mental well-being, our sleep quality, and our productivity, ultimately leading to an increase in mental illnesses. Therefore, this paper presents an ML-based digital health decision support system that utilizes one’s daily/weekly/monthly smartphone screen time as well as one’s behavioral data to predict the likelihood of suffering from a mental illness. Using user behavior analysis via machine learning and a digital twin–based prediction mechanism, our framework can analyze an individual’s current behavior to predict how this behavior may affect future mental health risk progression. The system will use varying metrics of daily behaviors (e.g., daily screen time, hours of sleep, stress levels, productivity levels, and well-being index) in conjunction with multiple classification models (e.g., KNN, SVM, Decision Tree, and Ensemble) to classify a user’s risk of developing a mental disorder. The ensemble model has achieved 90% accuracy and shown greater than 94% confidence in correctly classifying a case as high risk. The system will also provide a visual and quantitative assessment of 12 months of digital twin risk predictions, enabling individuals to proactively address and reduce the risk of developing a mental disorder. Finally, our digital health decision support system will generate automated clinical recommendations and prepare personalized health reports to help with early detection, increased awareness of digital well-being issues, and ongoing preventive mental health monitoring.