Finding Mirage Effect of Fraudulent Financial Statements Through Red Flags: A Comparative Analysis of Neural Network & Logistic Regression

Financial statement fraud (FSF) seems to be the order of the day nowadays. The aim of conducting nancial statement fraud is to deceive the investors and make them a fool by showing a mirage image to the public. Today's investor necessarily needs any tool or technique that can advocate for them to nd this mirage effect which wrongfully convinces them to invest their hard- earned savings. The techniques we are discussing here are two data mining techniques that are vitally used and gained paramount importance nowadays. These are Neural Network (NN) and Logistic Regression (LR). This study particularly demonstrates the superiority of these techniques which can, to a greater extent, accurately classify fraudulent and non-fraudulent companies based on information available in a company's nancial statement. The objectives of this study are also to nd red ags of fraudulent nancial statements and compare the accuracy levels of both techniques i.e. up to which extent they can classify the companies as fraud and non-fraud using the red ags as input vectors.

This study is conducted with a sample of 168 Indian companies identid by SFIO (Serious Fraud Investigation Office) and the PROWESS database. ⁱBased on past literature, 48 variables in the form of nancial ratios and corporate governance variables are found as predictive indicators of nancial statement fraud. Assuming the signiant variables as red ags and input vectors, the Multilayer Perceptron (MLP) method of Neural Network and Logistic Regression technique is applied using SPSS. After applying the methods, the study compares the Neural Network results with Logistic Regression (LR) output. As a result of this study, red ags are identid to nd the mirage effect. Both techniques are equally efficient in nding the mirage effect in fraudulent nancial statements and classifying fraud and non-fraud companies with up to 100% accuracy. A predictive equation through Logistic Regression is also formed. In this way, we can contribute to the fast development of society by producing an early warning system for investors, bankers, government, medical practitioners, and other professionals. ⁱⁱThe application of data mining techniques will surely prove to be one of the most important, practical, and fruitful endeavors in nance and other areas in the future decade.