Prediksi Hasil Putusan Pengadilan Pajak Kementerian Keuangan Menggunakan Light Gradient Boosting Machine (Light GBM)

Abstract

The increasing number of unresolved tax dispute cases in the Indonesian Tax Court, reaching 9,201 cases in 2024, highlights a significant challenge. To address this issue, the Ministry of Finance has implemented the e-tax court to enhance administrative and trial efficiency. However, additional tools are needed to assist judges in decision-making. This model is essential due to the necessity of adhering to deadlines in dispute resolution. This study develops a prediction model using historical tax court decision data categorized into three outcomes: 'Reject,' 'Fully Approve,' and 'Partially Approve.' The dataset includes 2,186 decisions with 14 case information attributes. The methods applied include data collection, preprocessing, classification, and evaluation with the optimal model scenario using an 80:10:10 split and 6-fold cross-validation. The test results indicate an accuracy of 81.7% and an F1 score of 81.5%. This indicates that the model performs quite well in predicting the target classes, although there are still some misclassifications that need to be addressed for further research.