Analysis of Indonesian Language Dataset for Tax Court Cases: Multiclass Classification of Court Verdicts

Abstract

Tax is an obligation that arises due to the existence of laws, creating a duty for citizens to contribute a certain portion of their income to the state. The Tax Court serves as a judicial authority for taxpayers seeking justice in tax disputes, handling various types of taxes daily. This paper analyzes an Indonesian language dataset of tax court cases, aiming to perform multiclass classification to predict court verdicts. The dataset undergoes preprocessing steps, while data augmentation using oversampling and label weighting techniques addresses class imbalance. Two models, bi-LSTM and IndoBERT, are utilized for classification. The research produced a final result of the model with 75.83% using the IndoBERT model. The results demonstrate the efficacy of both models in predicting court verdicts. This research has implications for predicting court conclusions with limited case details, providing valuable insights for legal decision-making processes. The findings contribute to legal data analysis, showcasing the potential of NLP techniques in understanding and predicting court outcomes, thus enhancing the efficiency of legal proceedings.