An Outlier Detection Method for Data Streams Based on Adaptive Clustering and Verification

This article introduces a novel outlier detection method for data streams named Adaptive Clustering and Outlier Verification (ACOV), which is based on adaptive clustering and outlier verification. ACOV is capable of distinguishing between false outliers induced by concept drift and true outliers. The ACOV framework principally comprises the Drift Detection (CDD) module, the Adaptive Clustering (AC) module, the Incremental Update (IU) module, and the Outlier Verification (OV) module. The CDD module is designed to differentiate between actual outliers and false positives caused by concept drift. The AC module first uses a diversity metric to adaptively select the value k and then uses the k-mRSR clustering method to detect outliers. The IU module is designed to accelerate the rate of outlier detection. The OV module creates a window overlap mechanism to verify whether a candidate outlier is indeed a true outlier, thus improving the precision of outlier detection. Experimental comparison with state-of-the-art methods shows that the proposed ACOV method exhibits commendable performance in terms of precision, recall, F1 score, and AUC metrics.