Data mining application to detect financial fraud in Indonesia's public companies

Abstract

Association of Certified Fraud Examiners explains that there are 3 types of occupational fraud: financial statement fraud, asset misappropriation and corruption. Among these three, financial statement fraud caused the biggest losses, which amounted to $ 1,000,000 in 2014. Financial statement has important role as an indicator of the success of a company, also for depicting the overall condition of the company, deciding company's stock price, and determining whether the company could be granted a loan or not. Given its important role, many cases of fraud occur. Audit activities are conducted to minimize losses, but the number of available auditors is limited, and the time required for traditional audit is quite long. Therefore, an effective model of financial fraud detection is needed to help auditors in analyzing financial statements. Data mining algorithms, support vector machine (SVM) and artificial neural network (ANN), were applied in this study. The results of this study give insight to the auditor that significant indicators in detecting financial fraud are profitability and efficiency. Feature selection improves SVM algorithm accuracy to 88.37%. ANN produces the highest accuracy, 90.97%, for data without feature selection.