KNN-SMOTE: An Innovative Resampling Technique Enhancing the Efficacy of Imbalanced Biomedical Classification

In recent years, the prevalence of imbalanced data has escalated within real-world problem domains, inducing substantial challenges that impede data analysis and exploration. This predicament is particularly pronounced within biomedical data classification tasks, wherein the imbalance quandary has garnered notable attention from the data science and machine learning research community. Various approaches, notably SMOTE and its variants, have been developed to address this concern and have demonstrated encouraging outcomes. Nonetheless, a comprehensive resolution to this issue remains elusive, with instances where the effectiveness of these methods falters or even diminishes classification performance. Consequently, this paper introduces a novel method named KNN-SMOTE, an enhancement of the Borderline-SMOTE technique. Empirical assessments conducted on five imbalanced benchmark datasets from the UCI Machine Learning Repository underscore the superior performance of our approach in terms of F-score, G-mean, and AUC in comparison to other methodologies.