Applied Research in Financial Reporting

Applied Research in Financial Reporting

Investigating the hybrid approach of feature selection methods with logistic regression and machine learning classification algorithms to improve the accuracy of earnings management prediction

Document Type : Original Article

Authors
Hassan Hassani
Esfandiar Malekian Kallehbasti
Yahya Kamyabi
Faculty of Economics and Administrative Sciences, University of Mazandaran, Babolsar, Iran
10.22034/arfr.2025.509803.2129
Abstract
Earnings management forecasting is an integral part of financial-economic analysis that helps shareholders, investors, creditors and outsiders to obtain high quality financial information of the company. The purpose of this research is to investigate and compare the performance of hybrid feature selection methods with classification algorithms of machine learning methods (including decision tree, k-nearest neighbor, deep learning and ensemble method of Adaboost- support vector machine) and logistic regression method to improve the accuracy of earnings management prediction. In this regard, By using feature selection methods based on relief and particle swarm optimization, earnings management prediction was discussed. In this research, 180 companies admitted to the Tehran Stock Exchange were selected as a statistical sample for the years 1389 to 1400. Also, to test the hypotheses, the criteria of average accuracy and type I and type ΙΙ errors were used. The results show that the performance of companies' earnings management forecasting methods based on relief-based feature selection model is better than the feature selection model based on particle swarm optimization. This result was confirmed in all methods of predicting earnings management. Also, the results indicate the superiority of machine learning methods over logistic regression. In addition, the results show that the earnings management prediction model created by combining the relief method and deep learning provides the best prediction performance with an average accuracy of 89.62%
Keywords
Accruals
feature selection
optimal predictor variables
Earnings management
machine learning

  • Receive Date 01 March 2025
  • Revise Date 05 September 2025
  • Accept Date 07 September 2025