Triyanna Widiyaningtyas, Hairani Hairani, Didik Dwi Prasetya, Utomo Pujianto, Wahyu Caesarendra
Imbalanced class distribution remains a significant challenge in healthcare data analysis, particularly in disease-related datasets where minority classes representing critical conditions such as diabetes are severely underrepresented. This disproportionate representation often results in biased predictive models that exhibit reduced sensitivity to minority classes, leading to suboptimal diagnostic accuracy and reduced generalizability. Imbalanced data can decrease the performance of classification methods and result in overfitting. SMOTE is a frequently used method for addressing data imbalance. A recent SMOTE variant considers only outliers to remove minority classes (data noise) without considering minority data neighboring majority classes, which are considered noise. This research aimed to modify SMOTE based on KNN filtering and a modification of Manhattan-based distance metrics to reduce the generation of noise data in minority classes and minimize overlap. The proposed method is called NR-Modified SMOTE and has several stages in balancing data: (i) filtering by removing minority classes close to majority classes (data noise) using the KNN method, and (ii) applying SMOTE oversampling with the modification of the Manhattan distance metric. Experiments were carried out on two health datasets, Pima and Haberman, with NR-Modified SMOTE and classification using Random Forest, SVM, and Naive Bayes using 10-fold cross-validation, where the proposed method led to better accuracy for all classifiers than NR-SMOTE without distance metric modifications. © 2025, Dr D. Pylarinos. All rights reserved.
Department of Electrical Engineering and Informatics, Faculty of Engineering, Universitas Negeri Malang, Malang, Indonesia; Department of Computer Science, Universitas Bumigora, Mataram, Indonesia; Department of Mechanical and Mechatronics Engineering, Faculty of Engineering and Science, Curtin University Malaysia, Malaysia
