What lies ahead developing a fraud prediction model: Application of artificial intelligence methods using firm-specific data and locational factors

• Main Authors: Apolinar, Jeneva Marielle B., Kung, Jim Ericson B., Ramirez, Jill Irish C., Rebadomia, Winona C.
• Format: text
• Language: English
• Published: Animo Repository 2014
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/12188
• Institution: De La Salle University

In light of the constant threat that fraudulent activities pose to various stakeholders, this study sought to develop a forecasting model that could predict the occurrence of fraud in companies based on publicly available financial and locational information, specifically: current ratio, total asset turnover, return on assets, debt to asset ratio, current asset to total asset ratio, corruption perception index (CPI) and gross domestic product (GDP). This study used logistic regression to analyze the independent variables in addition to creating a fraud prediction model, and the findings showed that all the variables, apart from total asset turnover and current asset to total asset ratio, were significant. Among them, current ratio, total asset turnover, debt to asset ratio, and current asset to total asset ratio have a positive effect on the probability of fraud occurrence while the return on assets, CPI, and GDP have a negative effect on fraud occurrence. Marginal effects (mfx) were used to measure the effect of a per unit increase/decrease in the independent variable on the dependent variable. Moreover, two artificial intelligence models using neural network and fuzzy logic were developed and compared to determine which is the most accurate and fit to be used by stakeholders. The results showed much promise, with the most accurate model having an accuracy rate of approximately 74% based on the F-score computed from the confusion matrix.