An Innovation of Machine Learning in Mental Health: Predicting Mood in Real-Time using Bio signals from Wearable Devices and Patterns of Mobile Phone Use

Abstract

Globally, mental health issues are on the rise, and mood assessment approaches rely on episodic self-reports and rare clinician tests. Traditional methods often miss everyday mood swings. A unique machine learning framework that integrates continuous biosignal data from wearable devices with passive behavioral cues from mobile phone use patterns predicts mood states in real time. HRV, EDA, skin temperature, and movement are captured via wearables to monitor autonomic nervous system activity connected to emotional regulation. Mobile phone usage logs also include conversation frequency, app usage, mobility, and sleep proxies. Advanced deep learning models that combine multimodal time-series data teach the system intricate temporal connections and baseline physiological and behavioral processes. We present a hierarchical model architecture for tailored feature extraction, temporal attention weighting, and mood state categorization across several emotional dimensions. Real-time predictions are verified against ecological momentary assessments obtained by smartphone prompts, exhibiting considerable accuracy improvements over baseline statistics and classical machine learning models. Results show that biosignal dynamics and mobile usage patterns can predict mood swings with high accuracy and context, enabling early mood disturbance identification and mental health interventions. This study shows that ubiquitous sensing and machine learning can make mental health care proactive and tailored.