MDBSCAN+: Enhancing MDBSCAN for Outlier Detection

Clustering is a vital technique in data analysis, but standard methods like DBSCAN may struggle with datasets that contain clusters of varying densities. MDBSCAN addresses these limitations by classifying data into low- and high-density regions before refining and merging clusters. However, MDBSCAN’s reliance on multiple hyperparameters complicates its outlier detection. Meanwhile, MS۲OD detects outliers in multi-density datasets through a scaled Minimum Spanning Tree (MST) approach but can misclassify smaller or sparsely located clusters as outliers. To overcome both MDBSCAN’s and MS۲OD’s shortcomings, the proposed MDBSCAN+ integrates the two approaches. It first detects potential outliers with MS۲OD, then applies MDBSCAN for clustering and evaluates outlier candidates via k-nearest neighbor distances. MDBSCAN+ effectively handles multi-density datasets and accurately identifies outliers with only one additional parameter (k), making it both efficient and user-friendly.