EGADB: An Enhanced Genetic Algorithm for Class Imbalance Problems

Class imbalance remains a critical challenge in machine learning, particularly in high stakes domains such as fraud detection and medical diagnosis, where minority class instances carry greater significance. Traditional classifiers often exhibit biased performance toward the majority class, leading to suboptimal detection of rare but important events. To address this challenge, this paper proposes an Enhanced Genetic Algorithm based Data Balancer (EGADB) that optimally resamples imbalanced datasets prior to classification. The proposed technique leverages genetic optimization principles to generate a balanced training set while preserving the intrinsic data distribution. The effectiveness of EGADB is evaluated using multiple base learners including Logistic Regression, Naive Bayes, and Bagging, as well as a developed voter based ensemble model. Experiments are conducted on the Credit Card Fraud Detection dataset and a breast cancer dataset, comparing classifier performance on unbalanced data and data balanced using EGADB. Performance is assessed using accuracy, precision, recall, F1 score, optimized precision, Matthews correlation coefficient, and balanced accuracy score. Experimental results demonstrate that EGADB significantly enhances minority class recognition and overall classification robustness across all evaluated models. The developed EGADB voter ensemble consistently achieves superior performance, attaining near perfect detection metrics on balanced datasets while maintaining stability across evaluation measures. These findings confirm that EGADB provides an effective and scalable solution for mitigating class imbalance and improving ensemble based predictive performance in imbalanced learning environments.