Smoothing Weight Distance to Solve Euclidean Distance Measurement Problems in K-Nearest Neighbor Algorithm
Guruh Putro Dirgantoro, Moch Arief Soeleman, Catur Supriyanto · 2021
In this paper we propose a k-nearest neighbors (kNN) classifier optimized, called swd-kNN, which can improve the performance of the original kNN classifier. kNN has a weakness in the distance measurement function using Euclidean, where each data attribute is considered the same. The attribute weighting obtained by the dummy data value generation method as a weight calculation, the performance of WKNN is proven to outperform the performance of kNN. swd-kNN integrates min-max normalization by integrating the WKNN weighting model which is used for parameter weighting in the feature weighting function at Euclidean distance. In this study using forest fire dataset from UCI Machine Learning Repository. Results of the accuracy measurement show that the performance of swd-kNN is better than the kNN and WKNN algorithms.