Local Outlier Reclassifier (LORec): a Method for Relocating Local Outliers Generated by K-means

The occurrence of local outliers produced by the K-means algorithm remains challenging since they affect the clustering performance due to misclassification. While current algorithms can identify the local outliers in the clusters, they do not relocate them to their correct clusters. Also, modifications of K-means addresses problems including time complexity, identifying the initial centroids and number of clusters, and preventing the occurrence of outliers in every cluster, but only a few focused on relocating the local outliers to their correct clusters. Hence, this paper introduces a Local Outlier Reclassifier (LORec) method capable of relocating local outliers generated by K-means. Results of the study using the three datasets show that the LORec integrated into K-means improved its clustering performance. The generation of false-positive instances is reduced by an average of 25.01%. Additionally, the integration yielded an average improvement of 8.50 % accuracy, 5.39% in the rand index, and 8.9% in f-measure. These results indicate that integrating LORec into K-means is an effective method for relocating local outliers to their correct clusters.