A Comprehensive Study of the Performances of Imbalanced Data Learning Methods with Different Optimization Techniques

Imbalanced classification is most important problem in machine learning, where the distribution of classes in the training dataset is skewed towards one class. So, insufficient representation of data in interested class (minority class) leads to misclassification by the canonical classifier, where learning by training datasets is biased towards the majority class. Several important applications are imbalanced in nature. Many different methods were proposed by the researchers to improve classification performance in an imbalanced dataset. The most common methods are resampling, cost-sensitive learning, ensemble methods, anomaly detection, etc. But there is a huge research gap in the prediction of an imbalanced or highly imbalanced dataset because the proposed methods are not assured about prediction accuracy. In cases of imbalanced datasets, hyperparameter tuning becomes more important to improve accuracy because a poorly tuned predictive model can easily neglect the interested classes (rare events), resulting in poor performance. This paper focused on comparative studies of the performance of different optimization methods with a proposed classifier for an imbalanced dataset.