Latifoğlu F., Turturova S., Orhanbulucu F.
PAMUKKALE UNIVERSITY JOURNAL OF ENGINEERING SCIENCES-PAMUKKALE UNIVERSITESI MUHENDISLIK BILIMLERI DERGISI, cilt.32, sa.6, ss.1265-1273, 2026 (ESCI, TRDizin)
Özet
Background and Objective: Major Depressive Disorder (MDD) is a significant mental health issue that negatively affects individuals' quality of life. In this study, an adaptive signal processing and machine learning-based approach was developed for the intelligent diagnosis of MDD through electroencephalogram (EEG) signals. Methods: The Variational Mode Decomposition (VMD), Empirical Mode Decomposition (EMD) and Empirical Wavelet Transform (EWT) methods were applied to EEG data from 26 MDD patients and 29 healthy control subjects using a limited number of EEG channels obtained from the frontal region, thereby decomposing the signals into sub-frequency bands. Frequency and statistical features were extracted from the raw and decomposed EEG signals. The LASSO algorithm was used for feature selection, and classification was performed using four different machine learning models. Results: The results obtained show that adaptive signal processing methods significantly improve performance compared to raw EEG analysis. In particular, the VMD & artificial neural network (ANN) combination emerged as the most successful model. This model demonstrated superior classification performance compared to other models, with 96.66% accuracy, 98.60% AUC, and 96.72% F1 score values. Furthermore, the decision mechanism of this model was examined in detail using SHAP analysis, and potential biomarkers that could guide clinical decision-making processes in MDD diagnosis were identified. Conclusion: In this context, the findings obtained in the proposed study significantly contribute to improving early diagnosis processes and developing automated diagnosis systems, providing a solid foundation for the development of computer-based diagnostic tools that will strengthen clinical decision support mechanisms.