Major depressive disorder (MDD) or depression is a complex and polygenic condition influenced by genetic, environmental, and psychosocial factors. Depression risk prediction and treatment outcomes remain a challenge due to multiple determinants. This work utilizes the ICBrainDB dataset to develop and evaluate gene-based depression prediction models utilizing 594 genetic samples and applied conventional machine learning techniques on the genetic biomarkers to build predictive models for early depression assessment. The approach integrates polygenic risk scoring and feature selection methods to enhance model interpretability and improve prediction accuracy. Comparative analysis between full and balanced datasets produced an understanding of how data distribution impacts performance. Notably, the proposed findings indicate that an artificial neural network achieved an 86% classification accuracy on the full dataset, while traditional machine learning techniques performed favorably on the balanced dataset. This suggests that neural network architectures may offer superior performance in capturing the intricate, nonlinear relationships inherent in genetic data even when faced with imbalanced datasets. This work contributes toward genetic analysis for psychiatric diagnostics and has the potential to automate the depression prediction tasks emphasizing the role of pharmacogenomics and computational modeling in mental health research.