Legal text classification in Turkey: A machine learning approach to divorce and zoning decisions

Abstract

The increasing volume of legal data in recent years requires integrating artificial intelligence (AI) techniques for efficient management and use. Critical challenges include classifying legal texts into specific fields or topics. This is crucial to advancing legal research and practice. This article aims to categorically classify Turkish court decisions, an area that has yet to be adequately researched before, compared to classification studies in international law texts. The study aims to contribute significantly to developing artificial intelligence-supported solutions to guide Turkish legal decisions by dividing legal texts into specific areas, thus increasing the efficiency and accessibility of the legal system. The study first created a data set consisting of divorce and zoning cases. Then, basic models were established with K-Nearest Neighbor (KNN), Support Vector Machines (SVM), Decision Trees (DT), and Random Forests (RF) algorithms to determine the algorithm that would classify the cases most effectively. Hyperparameter optimization was performed for each model to increase the Base Model performances. This process was supported by the 10-fold cross-validation method. Improved models were established with the hyperparameter values obtained as a result of optimization. As a result of comparative analysis, the SVM model had an impressive 90% accuracy rate in classifying legal texts. This result will significantly contribute to the development of intelligent legal systems by achieving significant success in classifying legal texts in Turkey.