Predictive and adaptive Drift Analysis on Decomposed Healthcare Claims using ART based Topological Clustering

Fraud in healthcare services dissipates funds that are important for improving the quality of life of people, thus enhancing the interest in predictive fraud analysis. The predictive analysis of fraudulent activity can be done by looking for unusual patterns in healthcare claims. However, unusual patterns may also occur due to sudden changes, isolated events, or concept drifts that frequently happen in healthcare which should not be considered fraud. Furthermore, analyzing drifts also supports predicting future trends and behaviors. In this study, we propose a novel approach, Drift Analysis on Decomposed Healthcare Claims (DADHC), to analyze the hidden patterns that hinder the performance of fraud prediction and detection. Our proposed model decomposes the series of healthcare claims into regular and irregular patterns using Psuedo Additive Decomposition (PAD) integrated with Simple Moving Average (SMA) smoothing technique. Then ART (Adaptive Resonance Theory) based Topological Clustering (TC) is used to analyze unusual patterns and identify the actual fraudulent activities. Our proposed model also incorporates correntropy based vigilance testing in ART to enhance adaptivity. Empirical evaluation on CMS Part B claims shows that our proposed approach has significantly improved detection accuracy compared to existing models due to the drift analysis.