An Intelligent Machine Learning Classifier System for Class Imbalance Learning and Improved Prediction Accuracy

Abstract

One of the major problems with machine learning is class imbalance, which occurs when the majority class is much larger than the minority one and the classifiers tend to learn the majority class rather than the minority ones, and hence make poor predictions in the latter scenario. This paper introduces an Intelligent Machine Learning Classifier System for Class Imbalance Learning and Improved Prediction Accuracy in a single framework that combines adaptive class balancing (ACCB), dynamic feature selection (DFS), hybrid ensemble learning (HEL) and confidence-based decision fusion (CFDF). Initially, data preprocessing is used to remove inconsistencies from the data and to normalize the distribution of features. To obtain a balanced learning, an Intelligent Adaptive Class Balancing Algorithm is used to generate representative minority samples and to reduce majority instances to be redundant. After that, a Dynamic Feature Importance Selection Algorithm determines the attributes that are discriminative and improve classification accuracy and decrease computational complexity. A Hybrid Ensemble Classification Algorithm using several machine learning classifiers is used to process the selected features to improve the robustness and generalization. Finally, a Confidence-Based Intelligent Prediction Algorithm is used to generate reliable prediction with the estimation of confidence. Experimental results on standard imbalanced datasets show that the accuracy, precision, recall, F1-score, G-Mean and ROC-AUC are always improved than the traditional machine learning methods. The proposed framework is able to effectively deal with the class imbalance, reduce the prediction bias, improve minority class detection and offer a scalable and interpretable solution for various real world intelligent decision support applications in different domains.