Using machine learning-based systems to help predict disengagement from the legal proceedings by women victims of intimate partner violence in Spain

Abstract

Intimate partner violence (IPV) is an actual social issue which poses a challenge in terms of prevention, legal action, and reporting the abuse once it has occurred. In this last case, out of the total of female victims that fill a complaint against their abuser and initiate the legal proceedings, a significant number withdraw from it for different reasons. In this field, it is interesting to detect the victims that disengage from the legal process so that professionals can intervene before it occurs. Previous studies have applied statistical models to use input variables and make a prediction of withdrawal. However, it has not been found in the literature any study that uses machine learning models to predict disengagement from the legal proceedings in IPV cases, which can be a better option to detect these events with a higher precision. Therefore, in this work, a novel application of machine learning techniques to predict the decision of victims of IPV to withdraw from prosecution is studied. For this purpose, three different ML algorithms have been optimized and tested with the original dataset to prove the great performance of ML models against non-linear input data. Once the best models have been obtained, explainable artificial intelligence (xAI) techniques have been applied to search for the most informative input features and reduce the original dataset to the most important variables. Finally, these results have been compared to those obtained in the previous work that used statistical techniques, and the set of most informative parameters has been combined with the variables of the previous study, showing that ML-based models have a better predictive accuracy in all cases and that by adding one new variable to the previous work' subset, the accuracy to detect withdrawal improves by 7.5%.