Investigating Hybrid Multiquadric–Gaussian Kernels in SVM for Depression Detection: A Preliminary Study

Accurate and reliable classification of depression is crucial for timely diagnosis, effective intervention, and improved patient outcomes. Traditional diagnostic methods for depression rely heavily on subjective assessments, which can introduce variability and inaccuracies. To overcome these limitations, advanced machine learning techniques, particularly Support Vector Machines (SVM), offer promising avenues for objective, precise classification. This research systematically investigates three newly proposed hybrid kernel strategies, Linear Combination, Multiplicative Combination, and Nested Combination, derived from Multiquadric (MQ-RBF) and Gaussian (GA-RBF) radial basis function kernels within the SVM framework. The empirical evaluations were conducted using two publicly available depression datasets encompassing diverse demographic, psychosocial, academic, and behavioral features. Experimental results clearly indicate the superior performance of the Nested Combination kernel, achieving classification accuracy of up to 88.05%, precision of 85.11%, recall of 94.12%, and an impressive ROC-AUC of 0.968. This kernel consistently outperformed baseline kernels and other hybrid forms across different training-test splits, demonstrating its robustness and adaptability. These promising outcomes underscore the practical significance and clinical potential of the Nested Combination kernel as a diagnostic support tool. However, future validation with larger and more diverse clinical datasets is recommended to further reinforce and generalize the kernel’s applicability in real-world depression classification scenarios.