Enhanced Kernel k-Nearest Neighbors Regression with Backward Feature Selection

This study presents an enhanced kernel <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{k}$</tex>-nearest neighbors regression method that incorporates backward feature selection to improve prediction accuracy. The Gaussian, triangular, and quartic kernels are of particular interest. Performance was evaluated using five distinct datasets, including high-dimensional data. The results demonstrate that the proposed method achieves lower root mean square error, lower mean absolute error, and higher coefficients of determination in most cases compared to traditional kernel KNN regression, indicating superior predictive capability and model fit.