NCBIS: A Novel Clustering-Based Approach for effective Instance Selection

Instance selection methods are key preprocessing techniques in data mining and machine learning, aimed at reducing dataset size, removing noise, and enhancing classification accuracy. However, many existing methods face limitations due to heuristic approaches, complex parameter tuning, and high computational demands. To address these issues, we propose a novel clustering-based instance selection approach (NCBIS) that leverages the intrinsic structure of data classes. NCBIS introduces a parameter-free, initialization-insensitive clustering algorithm to identify core instances, preserving dataset representativeness while minimizing manual intervention. Initially tested on small and medium-sized datasets, NCBIS demonstrates efficiency and resolves challenges like information loss and overfitting. Experiments on 15 benchmark datasets show that NCBIS outperforms state-of-the-art methods in accuracy, reduction rate, and computation time, making it a promising tool for data preprocessing and future research.