Fraud detection in financial statements using evolutionary computation based rule miners

Abstract

In this paper, we propose new rule based classifiers based on Firefly (FF) and Threshold Accepting (TA) Algorithms viz., Improved Firefly Miner, Threshold Accepting Miner, Hybridized Firefly-Threshold Accepting (FFTA) based Miner for classifying a company as fraudulent or non fraudulent with respect to their financial statements. We apply t-statistic based feature selection and investigate its impact on the results. FFTA and TA miners turned to be statistically similar. Both algorithms outperformed standard decision tree both in terms of sensitivity and the length of rules. © Springer International Publishing Switzerland 2015.