Sequential Cluster Extraction by Noise Clustering Based on Local Outlier Factor

Handling outliers is an important issue in machine learning, including clustering tasks. Outlier-resistant methods like Noise Clustering and DBSCAN are used to decrease outlier impact. Noise clustering based on the Local Outlier Factor (NCLOF), a combination of noise clustering and Local Outlier Factor, enhances clustering robustness against outliers. NCLOF adjusts dissimilarity based on LOF to improve clustering performance. Sequential cluster extraction based on noise clustering is a method of obtaining cluster partition while reducing the effect of outliers by extracting clusters one at a time. Also it has the feature that the number of clusters need not be determined in advance. This paper proposes an algorithm for automatically estimating clusters using sequential cluster extraction based on NCLOF. Sequential cluster extraction methods based on noise clustering and NCLOF were compared using artificial and benchmark datasets. The results of numerical experiments suggest the usefulness of the proposed sequential cluster extraction method.