Grammar-Based Multi-objective Genetic Programming with Token Competition and Its Applications in Financial Fraud Detection

In this study, we propose a new approach based on Grammar-based Genetic Programming (GBGP), token competition, multi-objective optimization, and ensemble learning for solving Financial Fraud Detection (FFD) problems. Token competition is a niching technique to maintain diversity among individuals. It can be used to adjust the objective values of each individual, and the individuals with similar objective values but different meanings are separated. Financial fraud is a serious problem that often produces destructive results in the world and it is exacerbating swiftly in many countries. It refers to many activities including credit card fraud, money laundering, insurance fraud, corporate fraud, etc. The major consequences of financial fraud are loss of billions of dollars each year, investor confidence, and corporate reputation. Therefore, a research area called FFD is obligatory, in order to prevent the destructive results caused by financial fraud. We comprehensively compare the proposed approach with Logistic Regression, Neural Networks, Support Vector Machine, Bayesian Networks, Decision Trees, AdaBoost, Bagging, and LogitBoost on four FFD datasets including two real-life datasets. The experimental results showed the effectiveness of the new approach. It outperforms existing data mining methods in different aspects.