In recent years, China has made great improvements in food quality and safety control, but with the increase in the scale of the food industry, the demand for inspections has increased. Moreover, food safety inspection data has appeared high-dimensional, complex and non-linear characteristics, and these features will lead to low utilization of quantitative analysis data, which directly affects the accuracy of the risk forecast model based on data. This study proposed a risk forecast model of food safety which integrated analytic hierarchy process and extreme gradient boosting tree algorithm based on food safety inspection data, and the integrated model was optimized and improved by food safety restricted indicators, so as to achieve more efficient and accurate food safety risk assessment. Based on this, the rice hazard detection data of 31 provinces across the country except Hong Kong, Macao and Taiwan were used as examples to elaborate on the use of the model. The result of the model test revealed that the risk forecast model had strong stability and high accuracy, which could provide certain theoretical basis and reference for the evaluation and decision-making of food safety regulatory authorities.