Objective To conduct a scoping review on risk prediction models for violent aggression behavior in patients with schizophrenia spectrum disorders， so as to inform future research and clinical decisionmaking. Methods Risk prediction models for violent aggression behavior in patients with schizophrenia spectrum disorders were systematically searched in 9 Chinese and English databases， including Chinese Biomedical Literature Data， China National Knowledge Infrastructure， VIP， WanFang Data， PubMed， Cochrane Library， CINAHL， Embase， and Web of Science. The search period was from January 1， 2014 to July 6， 2024. Data on modeling methods， validation and presentation formats， predictors， and predictive performance were extracted， and standardized reporting was conducted. Results A total of 18 articles were included， and the sample size of model construction ranged from 57 to 1 426. The incidence of violent aggression behavior in patients with schizophrenia spectrum disorders was from 24% to 79.70%. Logistic regression and various machine learning algorithm models were employed， with supervised machine learning method demonstrating the best performance in predicting such behavior. Common predictors included history of violent aggression behavior， negative symptoms， history of antipsychotic medication use， age， hospital stay， educational level and adherence. Conclusions Existing predictive models demonstrate some effectiveness in forecasting violent aggression behavior among individuals with schizophrenia spectrum disorders. In the context of artificial intelligence， multimodal clinical prediction models should be utilized to conduct accurate risk assessments of such behaviors and optimize the key influencing factors contributing to their occurrence， and thereby providing a basis for early clinical intervention decisions.