Abstract
Many real-life datasets suffer from class imbalance, where one or more classes are under-represented in the dataset, resulting in reduced classifier performance, with the expected decline in quality of procedures depending on the classification results, such as financial losses to businesses or inferior product quality. Improving classifier accuracy by handling class imbalance will positively impact classifier accuracy. In this study, we present a Probability-Based Synthetic Minority Oversampling Technique P-SMOTE to generate new examples for the minority class. Our proposed solution improves classifier accuracy by enhancing the oversampled examples through sampling the probability distributions present in the data. Results show improved performance over algorithms in the literature, with an average F-score of 0.821 over 13 datasets using 5 classifiers.