Detection of Frauds in Financial Reporting

Many researches have been done on annual reports to detect whether it is fraud or not by the analytical and empirical part of the report. Annual reports provide information on a company’s activities throughout a year. By analyzing the annual report, we can identify the condition of the company whether it is in crisis or operating perfectly. This research deals with the data that can be obtained from the reports’ text to determine the probability of being a fraudulent annual report. The verbal content of the report which determines the linguistic features are being analyzed using natural language processing tools to distinguish fraud financial reports from non-fraud financial reports. A set of 60 annual reports were taken for the study. Out of which 30 annual reports are labelled as fraud and the other 30 is labelled as non-fraud. The set of fraudulent companies were selected on the basis of a reporting case of fraudulency of another company or the same company in any other year of non-reporting of cases. The features are selected using a wrapper method search algorithm. A neural network model of MLP (Multi-Layer Perceptron) algorithm is used to classify the data with an accuracy of 85.1%. Classifiers like SVM (Support Vector Machines), Logistic Regression, Naïve Bayes and Random Forest algorithms were also used to identify the best classifier out of all the algorithms. Performance of all the techniques used in this paper are being analyzed and presented in terms of accuracy, precision, recall, F1 score, TN rate and FN rate.