Depression Analysis and Detection Using Machine Learning: Incorporating Gender Differences in a Comparative Study

Depression is a significant mental health problem and presents a challenge for the machine learning field in the detection of this illness. This study explores the automated depression classification, leveraging computational techniques to address this issue. The proposed approach performs spectrogram analysis and utilizes several machine learning methods, including SVM (Support Vector Machine), Random Forest, MLP (Multilayer Perceptron), and DepAudioNet. The research examines four datasets, i.e., DAIC-WOZ, EATD Corpus, D-Vlog, and EMU, which vary in linguistic backgrounds (English and Chinese), depression classification scales, and gender representation proportions. Feature extraction employed parameters such as formant-related, MFCCs (Mel Frequency Cepstral Coefficients), and jitter parameters. The innovation of this study lies in strategic enhancement, which involves incorporating a gender-specific perspective. This is achieved through the implementation of a tailored feature vector methodology. In most models, this approach led to measurable improvements: SVM improved by 1.43% (p = 0.044), MLP-CNN by 7.25% (p = 0.008), and Perceptron by 7.39% (p = 0.05). Such results underscore the necessity of integrating more personalized methods into creating machine learning algorithms for mental diagnostics.