GeometricSMOTE-Enhanced Deep Gaussian Mixture Models for Imbalanced Data Classification

Imbalanced data classification poses a challenge in machine learning due to limited minority class samples, impacting model performance. We propose an innovative solution merging Geometric Synthetic Minority Oversampling Technique (SMOTE) and Deep Gaussian Mixture Models (GMM) for improved classification. Geometric SMOTE generates synthetic samples by considering geometric relationships with nearest neighbors, enhancing Deep GMM training. Deep GMM captures complex data distributions and generates realistic samples, bridging minority-majority class gaps. Integrating Geometric SMOTE synthetic samples enables Deep GMM to learn representative data patterns, improving classification performance. Experiments on benchmarks exhibit the approach's effectiveness. GeometricSMOTE-enhanced Deep GMM notably enhances accuracy, precision, recall, and F1-score for the minority class. The approach remains robust across imbalanced ratios and dataset traits, a promising solution for real-world imbalanced data challenges. It enhances deep learning models' performance, offering valuable utility.