Application of Data Mining Technique for Fraud Detection in Health Insurance Scheme Using Knee-Point K-Means Algorithm

Healthcare delivery is one of the most important functions of government. This task is, however, beset with so many difficulties in a developing country like Nigeria. The application of Information Technology (IT) is one way to overcome several of the problems besetting the sector. This work, therefore, focuses on the application of some computer-based techniques that could help to properly target investment in the sector and also drastically reduce fraud in health insurance. The Nigerian Health Insurance Scheme (NHIS) was introduced by the Nigerian government to make healthcare affordable to all citizens, irrespective of economic situation or occupation. This scheme is, however, known to be beset by fraudulent claims from health practitioners within the system. To reduce or possibly eliminate this fraud, we also applied Knee-point K-means Clustering method, which is capable of detecting fraudulent claims from Health providers. Cluster-based outliers were examined. Health providers claims submitted to Health Maintenance Organization (HMO) were grouped into clusters. Claims with similar characteristics were grouped together. Clusters with small populations were flagged for further investigations. For clustering using two (2) attributes, six (6) clusters are formed. 74% of claims are clustered into cluster 2, 16% are in cluster 1, 2% are in cluster 3 and 4, 6% are in cluster 5 and 0% are in cluster 0 which have a membership of less than 1%. It can be difficult to get your clustering model correctly without determining the value of k clusters first; we were able to carve out some interesting information from the results on health insurance claims. The results from the data collected from an HMO in Lagos Nigeria show that the total number of claims identified as possible anomalies from cluster-based outliers is 7 in Nigeria health insurance using probability of 0.6 as the cutoff point.