Bipolar disorder, which is also known as manic depression and unipolar depression, are mental health conditions that cause unusual mood swings and can eventually affect a person’s energy and ability to function. A key symptom of both disorders is inactivity. Traditionally, diagnosis involves psychiatric assessments like the MADRS test, but early detection is crucial. Recent studies have shown that analyzing actigraph data from wearable devices can reliably diagnose these disorders, offering a non-invasive and continuous monitoring option. This paper explores using near miss under-sampling and class weighting strategies to implement Random Forest, LSTM, and XGBoost algorithms. Among these, Random Forest achieved the finest performance, with 99% accuracy, precision, recall, and F1 score, highlighting its potential as a robust device for the initial and accurate detection of mood disorders.