According to the National Institute of Mental Health, Major Depressive Disorder affected an estimated 21.0 million American adults in 2020, which represents 8.4% of the U.S. population aged 18 or older in a given year. Even though the percentage is substantial, it reflects only the diagnosed cases. Most depression cases remain undiagnosed and thus untreated. Real-time monitoring of physiological indicators of depression using wearable health monitoring devices can help increase the chances of early detection and eventual treatment. In this research, various Artificial Intelligence algorithms are developed to look for signs of stress and anomalies in activity patterns from the data captured by wearable health devices. The Random Forest algorithm performed well in detecting depression from users' activity levels, while the K-Nearest Neighbours algorithm detected stress, one of the key indicators of depression, with an accuracy of 96.2% from Heart Rate variability. This research takes advantage of real-time access to one's physiological data to minimize the number of undiagnosed depression cases.

Stress, a pervasive issue in today's world, significantly impacts mental and physical health. With approximately 20% of adults experiencing mental health disorders each year, and depression affecting over 260 million people globally, effective stress manage- ment is crucial. Chronic stress is also associated with physical ailments, including cardiovascular diseases, the leading global cause of death. In parallel, the wearable device market has witnessed substantial growth, with over 300 million units sold worldwide in 2021. These devices offer a unique opportunity to collect physiological and behavioral data, making them invaluable for stress detection and prediction. Combining artificial intelligence (AI) and wearables, this paper explores their accuracy in capturing vital data and the potential to revolutionize stress management, ultimately improving mental and physical well-being. In this paper, the ensemble-based classifier known as the Histogram-based Gradient Boosting Classifier detected stress from heart rates with an accuracy of 73%. This establishes the independent nature of Heart Rates for stress prediction, emphasizing that there is no need for sophisticated ECG and HRV readings. This opens up the possibility of using sensors as used in wearable devices to effectively detect stress in our day-to-day lives.