Leveraging Digital Data and Soft Computing Techniques for Advanced Mental Health Analysis and Intervention Strategies

Here we look at an LSTM-SVM algorithm that leverages data on movement patterns from actigraph sensors to recognize sadness. Incorporating data from statistics such as mean, standard deviation, and zero-activity fraction, this framework combines high-order LSTM characteristics with other information. Training and testing the algorithm are done using the Depresjon information set, which contains actigraphy information from regular control people and patients with depression. In order to prepare the data for the extraction of features by the LSTM, it was resampled to 24-hour periods and then framed. When paired with the statistical data, such characteristics were fed into SVM for categorization. The model’s success rate was $95.57 \%$ and its F-measure was 0.955, outperforming comparable deep learning techniques like LSTM and CNN. Sliding doors were the game-changer that raised efficiency by $3.65 \%$; however, adding discriminatory statistical features further improved the algorithm’s accuracy by $1.08 \%$. The results validate the suggested hybrid model’s value in improving depressive disorders detection accuracy and provide credence to its applicability in psychological monitoring with respect to sensors worn. Compared to other deep learning approaches, this strategy produces the best identification model.