Taxpayer compliance classification using C4.5, SVM, KNN, Naive Bayes and MLP

Tax revenue has a very important role to fund the State's finances. In order for the optimal tax revenue, the tax authorities must perform tax supervision to the taxpayers optimally. By using the self-assessment taxation system that is taxpayers calculation, pay and report their own tax obligations added with the data of other parties will create a very large data. Therefore, the tax authorities are required to immediately know the taxpayer non-compliance for further audit. This research uses the classification algorithm C4.5, SVM (Support Vector Machine), KNN (K-Nearest Neighbor), Naive Bayes and MLP (Multilayer Perceptron) to classify the level of taxpayer compliance with four goals that are corporate taxpayers comply formally and materially required, corporate taxpayers comply formally required, corporate taxpayers comply materially required and corporate taxpayers not comply formally and materially required. The classification results of each algorithm are compared and the best algorithm chosen based on criteria F-Score, Accuracy and Time taken to build the model by using fuzzy TOPSIS method. The final result shows that C4.5 algorithm is the best algorithm to classify taxpayer compliance level compared to other algorithms.