Mental health disorders are a growing global concern, yet traditional diagnostic methods remain largely subjective and episodic, failing to account for real-time fluctuations in a patient’s condition. However, recent advances in sensor technology and machine learning (ML) enable continuous and objective mental health monitoring. This study aims to develop and validate ML models that integrate different sensor data, including physiological, behavioral, and environmental signals, and ML algorithms to enhance the accuracy and robustness of real-time mental health monitoring methods. Data were collected using wearable devices and smartphones, capturing health-related data such as heart rate variability and physical activity. ML models, including support vector machines, random forests, and neural networks, were trained to conduct data preprocessing, sensor data fusion, and hyperparameter optimization. Random forest outperformed other algorithms in terms of accuracy, precision, recall, and F1-scores of its prediction results. Adding regularization enhanced the performance of neural networks, and data fusion from multiple sensors is far more effective than using single-sensor data for diagnosing mental health disorders. By providing a scalable, real-time system that leverages multiple sensor data and ML optimization, challenges in mental health monitoring, such as data heterogeneity and missing values, are addressed, and the deployment of the diagnosis system in clinical and everyday life becomes easier. The developed system enables the continuous, accurate care and support for people with mental health disorders. By integrating additional datasets and privacy protection measures, the developed system further enhances its effectiveness and accessibility.