A Mental Health Performance Assessment using Support Vector Machine

Abstract

In this paper, the support vector machine (SVM) was used to assess mental health performance using the Open Sourcing Mental Illness (OSMI) in Tech Survey 2019 dataset. To improve the SVM’s performance, data pre-processing was performed using feature scaling, and an autoencoder was utilized as a feature representation for classification tasks. Different combinations of kernel types and gamma values were used with the SVM for performance assessment. The kernel types used included polynomial, sigmoid, radial basis function (RBF), Bessel, and ANOVA. The findings indicated that in the case of the RBF kernel, SVM outperformed other kernels. The average variation in accuracy with different split ratios is approximately between 91%-95%. The minor variations observed in accuracy across different split ratios suggest that the model is robust and can generalize well to new data. It shows the effectiveness of the approach in modeling complex relationships between input features and output labels. This study also highlights the importance of appropriate parameter tuning for the optimal performance.