O Uso da Mineração de Processos na Análise do Tempo das Movimentações Processuais

Abstract

The time required to finish legal proceedings impacts the economy, investor confidence in a country and the lives of the population. In Brazil, principles of good practice provide guidelines regarding the reasonable duration of a lawsuit. However, machine learning models can accurately predict total procedural time and provide insights into possible causes of inefficiency. Literature works provide solutions in this sense, but do not consider the procedural movements carried out in the process during its execution. This work uses machine learning models trained with characteristics based on the sequence of procedural movements and characteristics of the court responsible for the process. It thus investigates the ability of the models for the task, in addition to considering the importance of the features involved. There were used lawsuits from multiple northeast regional labor courts in Brazil (TRE). The models were able to predict the total procedural time with high accuracy in which the Light Gradient Boosting Machine (LGBM) obtained an average R² of 0.9. The results suggest the efficiency of the approach and the success of process mining techniques for the task.