Machine Learning Models for Speech-Based Depression Screening

Mental health issues significantly affect society, requiring long-term treatment that is both costly and impacts personal relationships. This research paper proposes a speech-based method to screen for depression risk as an accessible alternative. The research included five machine learning models-Logistic Regression, Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Naive Bayes, and Linear Discriminant Analysis (LDA)-were used to analyze speech data from 25 participants with varying mental health conditions. Three data preparation techniques-Leave-One-Out Cross-Validation, SMOTE, and raw data-were applied to optimize the models. Results showed that SVM with SMOTE achieved the highest accuracy (98 %), emphasizing the importance of effective data preparation. This research demonstrates the potential of using speech analysis as a cost-effective tool for early mental health screening and provides a foundation for practical applications in healthcare and beyond.