Machine Learning for Violence Risk Assessment Using Dutch Clinical Notes

Violence risk assessment in psychiatric institutions enables interventions to avoid violence incidents. Clinical notes written by practitioners and available electronic health records are valuable resources capturing unique information, but seldom used their full potential. We explore conventional deep machine learning methods assess patients using practitioner notes. The performance of our best models is comparable the currently questionnaire-based method, with an area under Receiver Operating Characteristic curve approximately 0.8. find that deep-learning model BERTje performs worse than methods. also evaluate data classifiers understand better. This particularly important for applicability evaluated new data, great interest practitioners, due increased availability format.