Predicting the Amount of Compensation for Harm Awarded by Courts Using Machine-Learning Algorithms

Abstract

Abstract The present study aims to explain and predict the monetary amount awarded by courts as compensation for harm suffered. A set of machine-learning algorithms was applied to a sample of decisions handed down by the Polish common courts. The methodology involved two steps: identification of words and phrases whose counts or frequencies affect the amounts adjudicated with LASSO regression and expert assessment, then applying OLS, again LASSO, random forests and XGBoost algorithms, as well as a BERT approach to make predictions. Finally, an in-depth analysis was undertaken on the influence of individual words and phrases on the amount awarded. The results demonstrate that the size of awards is most strongly influenced by the type of injury suffered, the specifics of treatment, and the family relationship between the harmed party and the claimant. At the same time, higher values are awarded when compensation for material damage and compensation for harm suffered are claimed together or when the claim is extended after it was filed.