NSO Natural Selection Optimization for Adaptive k-Nearest Neighbor Imputation

چکیده مقاله

The optimization of parameters in missing data imputation methods is crucial for the performance of classification models. This study addresses the challenges of optimizing parameters in iterative k-nearest neighbor (kNN) imputation methods. Despite kNN imputation's effectiveness, its performance is highly dependent on selecting appropriate parameters such as the number of neighbors (k) and distance metrics. Traditional optimization techniques, like grid search and Bayesian methods, often struggle with scalability and adaptability, especially in high-dimensional datasets. To overcome these limitations, we propose a Natural Selection Optimization (NSO) approach that mimics natural selection to dynamically tune kNN imputation parameters. NSO treats the parameter space as an evolving ecosystem, where parameter combinations compete for survival based on imputation accuracy and resulting classification performance. Through iterative generations, NSO adapts parameters in real-time, tailored to the dataset's characteristics. Our method was evaluated on diverse datasets, including medical diagnosis and credit scoring, showing significant improvements in imputation quality and classification metrics such as accuracy and F1-score. Compared to traditional methods, NSO demonstrates superior scalability and adaptability to varying missingness patterns, enhancing kNN imputation's effectiveness and opening new avenues for nature-inspired algorithms in data preprocessing tasks