Voice-Based Automated Detection of Mental Health Considering Comparative Performances of Multiple Machine Learning Models

Depression and anxiety significantly threaten well- being, with extreme consequences. The rise in mental health conditions needs effective methods of detection and treatment. Machine Learning (ML) based models greatly facilitate various aspects of mental healthcare. This work addresses the effectiveness of the voice and its features in acknowledging mental health states. Voice analysis removes the bias inherent in self- reported psychological assessments. We use multiple spectral and chroma-related features to predict depression and anxiety. Further, forward feature selection, decision tree, and calculation of Fisher's score are performed to enhance the performance of the model and identify the most relevant features like spectral flux (SF), zero crossing rate (ZCR), chroma vector (CV), and standard deviation pitch (SDP). This work uses various ML models, such as support vector machine (SVM), multi-layer perceptron (MLP), XGBoost (XGB), K-nearest neighbour (KNN), decision tree (DT), random forest (RF), logistic regression (LR), AdaBoost (AB), and Naive Bayes (NB), for predicting and finding the model that best fits our data with an accuracy of 97%. From the result, we observe that the extracted voice features serve as markers for mental health conditions such as depression and anxiety and offer in the detection and monitoring of the disorders.