A Systematic Survey on Machine Learning Classifier Systems to Address Class Imbalance

Class imbalance is a significant obstacle in machine learning, causing biased model results and constraining generalization abilities. This systematic survey reviews machine learning classifier systems designed to mitigate class imbalance challenges. The objective is to analyze the effectiveness of various techniques in overcoming imbalance hurdles. The survey defines class imbalance and its implications for model training and evaluation. It categorizes approaches into data-level, algorithm-level, and hybrid methods, including oversampling, undersampling, cost-sensitive learning, ensembles, and deep learning. Each category is explored to provide insights into strengths, limitations, and applicability. Evaluation metrics tailored for imbalanced datasets are scrutinized, emphasizing the importance of metrics prioritizing minority class performance. Challenges in evaluating classifier systems in imbalanced settings are discussed, with guidance on metric selection. Recent advancements and trends in handling class imbalance, such as active learning and interpretable methods, are highlighted. Key challenges like scalability, interpretability, and algorithmic robustness are identified, with discussions on open research directions. This survey aims to offer a comprehensive understanding of classifier systems for class imbalance, aiding researchers and practitioners. By synthesizing recent literature and analyzing imbalance challenges, this study contributes to advancing knowledge in this vital area of machine learning