Author Identifier (ORCID)
Abstract
Diabetes is a global health issue, and machine learning has become crucial for early detection. The k-Nearest Neighbors (k-NN) machine learning algorithm is widely used because it is simple and easy to implement; however, its performance is sensitive to feature scale differences, class imbalance, and the assumption of uniform decision-making, as commonly found in clinical datasets such as the Pima Indians Diabetes (PID) dataset. Many studies have been conducted to address these issues; however, very few have systematically evaluated the combined effects of preprocessing and distance-based decision-making methods on k-NN performance. This study evaluates the combined effects of preprocessing strategies and distance-based decision-making methods on k-NN performance. Two preprocessing methods, namely normalization and resampling, were evaluated. The normalization methods evaluated are MinMax and Z-Score, while the resampling methods evaluated are SMOTE, ADASYN, and Random Undersampling. Two weighted decision-making schemes, namely Dual Weighted and Gaussian Weighted Voting, were also investigated. Performance was evaluated using Accuracy, Precision, Recall, Specificity, F1-score, AUC, and execution time analysis. The results show that Z-Score normalization combined with SMOTE improves the accuracy of the baseline k-NN from 74.48% to 84.10% and increases recall from 54.90% to 90.80%. Incorporating Dual Weighted Voting achieves the highest overall performance, with 84.20% accuracy, 91.80% recall, 85.32% F1-score, and an AUC of 0.85, while maintaining low computational cost. However, the additional improvement over the best preprocessing configuration is not statistically significant. These findings indicate that performance gains are primarily driven by normalization and resampling strategies, while distance-sensitive voting provides marginal additional refinement for classical k-NN in imbalanced medical classification tasks.
Keywords
diabetes detection, k-nearest neighbors, normalization, resampling, weighted voting
Document Type
Journal Article
Date of Publication
6-30-2026
E-ISSN
2716621X
Volume
7
Issue
2
Publication Title
Journal of Soft Computing and Data Mining
Publisher
Penerbit UTHM
School
School of Business and Law
RAS ID
101691
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License.
First Page
388
Last Page
402
Recommended Citation
Agustiyar, Isnanto, R. R., Warsito, B., Wibowo, A., & Jie, F. (2026). Analysis of normalization, resampling, and weighted voting to improve K-NN performance in diabetes disease detection. Journal of Soft Computing and Data Mining, 7(2). https://doi.org/10.30880/jscdm.2026.07.02.021