Predicting the Category and the Length of Punishment in Indonesian Courts Based on Previous Court Decision Documents

Among the sources of legal considerations are judges’ previous decisions regarding similar cases that are archived in court decision documents. However, due to the increasing number of court decision documents, it is difficult to find relevant information, such as the category and the length of punishment for similar legal cases. This study presents predictions of first-level judicial decisions by utilizing a collection of Indonesian court decision documents. We propose using multi-level learning, namely, CNN+attention, using decision document sections as features to predict the category and the length of punishment in Indonesian courts. Our results demonstrate that the decision document sections that strongly affected the accuracy of the prediction model were prosecution history, facts, legal facts, and legal considerations. The prediction of the punishment category shows that the CNN+attention model achieved better accuracy than other deep learning models, such as CNN, LSTM, BiLSTM, LSTM+attention, and BiLSTM+attention, by up to 28.18%. The superiority of the CNN+attention model is also shown to predict the punishment length, with the best result being achieved using the ‘year’ time unit.