A threefold-ensemble k-nearest neighbor algorithm
Hassan I. Abdalla, Aneela Altaf, Ali Abdullah Hamzah · International Journal of Computers and Applications · 2025
Implementing the k-nearest neighbor classifiers is both conceptually and practically easy. However, kNN performance still suffers from the sensitivity to the selection of k-value and the incapability to generate competitive results over the unbalanced datasets. We therefore offer an ensemble-based, effective k-nearest neighbor classifier because of the ensemble learning’s resilience to class imbalance. The general KNN, weighted KNN, and local mean KNN are meticulously merged into a single ensemble KNN classifier in order to incorporate the ensemble weights from these classifiers. Given that these three kNNs are combined to create the final model, this kNN is called the Threefold-Ensemble K-Nearest Neighbor (TEkNN). The effectiveness of the proposed kNN model has been comprehensively assessed against five state-of-the-art kNN models and four machine learning models over 14 datasets from the University of California using the evaluation metrics (accuracy, F1, ROC, and MAE). The results illustrate that the TEkNN is a promising classifier across all evaluation metrics, attesting to its usability with high precision in other domains in which class imbalance is dominantly inherent.