This project was conducted during my final year in college as my thesis as well as being a research project for ClaimVantage. The aim of the project is described in the abstract attached. The project was developed as a J2EE project deployed on Tomcat with a AngularJS front end. The J2EE backend hooked upto the a Salesforce org where ClaimVantage have deployed there application.
The aim of this project was to conduct research in the area of fraud detection in the health insurance sector with the objectives of acquiring expert knowledge of indicators and patterns used to detect fraud in insurance claims and to develop an application incorporating this acquired knowledge which would allow the automated processing of claims using aspects of this knowledge but also allow an expert to use selected components and patterns to and deploy their own customized analysis. Research suggested that adopting an Artificial Intelligence (AI) approach could be used for this purpose. The most commonly used AI approach used in the healthcare sector is to adopt rule-based expert systems. This approach was therefore adopted in this project.
Before the application could be developed, a knowledge acquisition process had to be designed and implemented and the resulting knowledge analyzed to identify the core knowledge concepts used, and the relationships between these, by fraud analysts to detect fraudulent claims from which a set of global rules for use in the application were identified. This involved conducting a review of relevant literature to identify what types of fraud exist in the healthcare sector, the types of indicators and is commonly used and how these can be combined into patterns. From this a survey was developed and sent to a set of expert claims and fraud analyst who currently use the ClaimVantage automated claims processing software to identify expert views of which indicators are most important when attempting to identity fraud. The most highly ranked indicators were translated into rule components in the expert system development. Statistical analysis was conducted on the responses to the survey to identify correlations between indicators from which patterns or rule sets were developed for use in the expert system.
The expert system developed can detect fraud by running all claims currently in the system against the common patterns of fraud detection captured in the knowledge base created. The claims data is pulled from the ClaimVantage automated claims processing software. Further, the expert system allows analysts to configure personalized packages of rules, which represent their own fraud detection preferences, from the component rules identified, which they can store for future use or extension. The expert system was developed as a web application using the Drools rule engine to implement the required knowledge base, wrapped in a java EE project that will allow users to build powerful rule sets to run against the claims.
It is impossible to truly evaluate how the application would work in the wild without the access to a real world data set, which is impossible to achieve for a student project given the confidential nature of the data involved. However, it has been a success in terms of acquiring the knowledge and the wrapping of this knowledge in a custom developed expert system shell to run against the claims data stored in the ClaimVantage application and this success has been established by using a process of review of both the application development and the application itself by analysts and developers involved in the development of claims processing software.