This paper is about a system which applies a modified K - Means algorithm[12] to flag out suspicious claims for further scrutiny has been developed. The Java programming Language and my SQL database tools were used. The K - Means algorithm is well known for its efficiency in clustering large data sets. However, a major limitation of this al gorithm is that it works only with numeric values, thus the method cannot be used to cluster real - wo rld data containing categorical values. To counter this, data sets were converted to numeric data whereby ailments were listed and matched with patients. The pre sence of the ailment was represented by a one (1) and the absence was represented by a zero ( 0). To get the data, a total of 15 insurance companies in Kenya out of 31 were randomly selected and a pre - tested questionnaire was used to collect data. 15 insurance companies out of 31 is close to 50%, which is a very good representative of the entire population. 67 % of the respondents indicated that the people involved in the processing of claims were billing for services that were not rendered. The results also showed that all the companies had internal control mechanisms to address the problem and 47% of the respondents said the internal controls were not efficient. 87% of the respondents indicated that the common member fraud cases involved m membership substitution including card abuse.