A Dataset-Driven Parameter Tuning Approach for Enhanced K-Nearest Neighbour Algorithm Performance

The number of Neighbours (k) and distance measure (DM) are widely modified for improved kNN performance. This work investigates the joint effect these parameters in conjunction with dataset characteristics (DC) on Euclidean; Chebychev; Manhattan; Minkowski; Filtered distances, eleven k values, four DC, were systematically selected parameter tuning experiments. Each experiment had 20 iterations, 10-fold cross-validation method thirty-three randomly datasets from UCI repository. From results, average root mean squared error is significantly affected by type task (p<0.05, 14.53% variability effect), while DC collectively caused 74.54% change RMSE DM accumulated least 25.4%. interaction k, resulted DM='Minkowski', 3≤k≤20, 7≤target dimension ≤9, sample size (SS) >9000, as optimal performance pattern classification tasks. For regression problems, experimental configuration should be7000≤SS≤9000; 4≤number attributes ≤6, = 'Filtered'. performed most influential determinant, followed DM. variation accuracy resulting changes values only occurs chance, it does not depict any consistent pattern, its value other yielded a statistically insignificant (p>0.5). As further work, discovered patterns would serve standard reference comparative analytics algorithms.