<![CDATA[This study aims to detect potential tax evasion of Article 21 Income Tax in filling out the Annual Tax Return of individual taxpayers using artificial intelligence such as machine learning in the form of an artificial neural network (ANN) in the form of a multilayer perceptron (MLP). The number of data is 3,000,000 which is built from aggregate data of individual taxpayers provided by the Directorate of Tax Data and Information using the Monte Carlo method which is then separated into three (3) data frames, namely 1770, 1770S, 1770SS through data pre-processing. The multilayer perceptron (MLP) model is built using the backpropagation algorithm with a supervised learning approach. The activation functions used in the hidden layer and output layer are Tanh and Sigmoid, respectively. Dimension reductionality uses Principal Component Analysis (PCA) so that the model can work efficiently. The results show that there is potential for tax evasion of Article 21 Income Tax detected by the multilayer perceptron (MLP) model on Gross Income, Net Income, Taxable Income, Assets, Debts, and Reporting Status. Evaluation with cross-validation shows that the model detects with an average precision of 83.27% and an average accuracy of 75.19% and an average ROC-AUC evaluation of 76% on the three data frames which implies that the artificial neural network (ANN) model is adequate because it has successfully passed the threshold of 0.5 in detecting potential tax evasion of Article 21 Income Tax in Personal Income Tax Return in Indonesia.]]>
Copyrights © 2026
