Prediction of earnings manipulation on Malaysian listed firms: A comparison between linear and tree-based machine learning

Predicting the earning manipulation is an inseparable part of financial-economic analysis, helping shareholders, investors, creditors and outsiders acquire high quality of firm�s financial information. Thus, the aim of the paper is to compare the earnings manipulation prediction models developed by using two types of machine learning algorithms; linear and tree categories. The linear based machine learning are Logistic Regression and Generalized Linear Model while the tree based are Decision Tree and Random Forest. All of the algorithms were tested on dataset of earnings manipulation among 1874 firm-year observations of firms listed on Bursa Malaysia . The results indicate that the performances of the two kinds of machine learning is not extremely different except with the Decision Tree. Furthermore, the most outperformed algorithm has been presented by the linear based machine learning, which produced the best accuracy in the shortest total time completion. All the models present better ability in detecting the false cases of earnings manipulation rather than the true cases mainly from the tree based machine learning. © 2021 Sociedad Argentina de Genetica. All rights reserved.