Extraction and Analysis of Multi-modal Dataset for Mental Disorder Detection

 The prevalence of mental disorders such as major depressive disorder (MDD) necessitates the development of objective, multimodal biomarkers for diagnosis and treatment monitoring. In this study, a comprehensive analysis integrating electroencephalography (EEG), behavioral, and audio data to explore neural and affective signatures of depression presented. Inspired by recent initiatives like the MODMA dataset—which provides an open, multimodal resource for mental-disorder analysis—EEG data during both a dot-probe task and resting state using high-density (128-channel) and wearable (3-channel) systems acquired. Behavioral responses extracted from E-Prime outputs and acoustic features from audio recordings further enriched the dataset. Standard preprocessing methods, including band-pass filtering and independent component analysis (ICA) for artifact removal, were applied to ensure high data quality. From the cleaned EEG signals, spectral power in canonical 
frequency bands (delta, theta, alpha, beta, gamma) using Welch’s method computed. These features, in combination with behavioral and clinical data, were used to build a logistic regression classifier, which was evaluated using cross-validation and interpreted with SHAP analysis to reveal key feature contributions. Unsupervised techniques such as principal component analysis (PCA) and k-means clustering further uncovered latent patterns within the integrated data. Findings underscore the potential of multimodal fusion to yield robust neurophysiological markers that, when combined with behavioral and audio metrics, may  improve the objectivity and efficiency of depression diagnosis. This integrated framework sets the stage for future investigations into scalable, data-driven approaches to mental health assessment.