Comparison of the Performance of the Water Cycle Algorithm with Metaheuristic and Machine Learning Methods in Predicting Unethical Behaviors and Financial Fraud

Introduction: Unethical financial behaviors and financial statement fraud are among the fundamental challenges of accounting and auditing systems, leading to the misallocation of resources, reduced investor confidence, and financial crises. The purpose of this study is to compare the performance of the Water Cycle Algorithm with six other methods, including the Genetic Algorithm, Firefly Algorithm, Artificial Bee Colony Algorithm, Logistic Regression, Support Vector Machine, and Decision Tree, in predicting fraudulent financial statements and unethical financial behaviors. Material and Methods: The statistical population consisted of all companies listed on the Tehran Stock Exchange during the period ۲۰۱۶–۲۰۲۳. The final sample included ۳۰ companies and ۲۱۸ firm-year observations, of which ۲۸ observations (۱۲.۸۳%) were classified as fraudulent firms and ۱۹۰ observations (۸۷.۱۷%) as non-fraudulent firms. Using the Shannon entropy method, ۱۵ significant variables were selected from an initial set of ۴۲ variables. Subsequently, seven algorithms with optimized parameter settings were independently executed ۳۰ times each. The performance of the algorithms was evaluated using Accuracy, Precision, Recall, F۱-score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). The nonparametric Wilcoxon signed-rank test was employed to examine the statistical significance of the differences among the algorithms. Results: The Water Cycle Algorithm achieved the best performance among all methods, with an F۱-score of ۰.۸۶۹ and an AUC of ۰.۹۲۳. It was followed by the Genetic Algorithm (F۱ = ۰.۸۲۵, AUC = ۰.۸۸۷), Artificial Bee Colony Algorithm (F۱ = ۰.۸۱۸, AUC = ۰.۸۷۹), Firefly Algorithm (F۱ = ۰.۸۰۸, AUC = ۰.۸۷۱), Support Vector Machine (F۱ = ۰.۷۹۹, AUC = ۰.۸۶۲), Logistic Regression (F۱ = ۰.۷۸۲, AUC = ۰.۸۴۶), and Decision Tree (F۱ = ۰.۷۷۰, AUC = ۰.۸۳۱). The results of the Wilcoxon test indicated that the differences between the Water Cycle Algorithm and all other methods were statistically significant (p-value < ۰.۰۵ for all comparisons). Furthermore, the Water Cycle Algorithm, with a standard deviation of ۰.۰۲۵, demonstrated the highest stability among the algorithms examined. Conclusion: The Water Cycle Algorithm significantly outperforms other metaheuristic and machine learning methods in predicting unethical financial behaviors and fraudulent financial statements. Therefore, it can serve as an effective tool for auditors, financial analysts, and regulatory authorities.

کلیدواژه ها:

Unethical Financial Behaviors ، Financial Fraud ، Water Cycle Algorithm