In areas with higher latitudes, ice sheets, ice jams or ice dams are often formed in rivers to overtop levees due to lower temperatures in winter, which can cause sizeable social and economic losses, as well as countless injuries and death. This study presents the typical ice flood area of the Inner Mongolia section of the Yellow River as the research object. On the basis of the hazard system theory and comprehensive analysis of ice flood risk factors, including the hazard-inducing factors and hazard-pregnant environments, as well as the vulnerability of the hazard-bearing bodies, we selected eight risk assessment indicators to construct an ice flood hazard risk assessment model based on random forest (RF) algorithm. The three hazard-inducing factors consist of: maximum submergence depth, maximum flood velocity and maximum submergence duration, which were derived from the river-flood ice flood backwater burst submergence coupling model; while the three hazard-pregnant environments are: topographic elevation, terrain gradient and the distance from the river. The two hazard-bearing bodies include: population density and GDP density. The modeling results show that compared with the risk assessment model of K-nearest neighbor (KNN) algorithm, both the index Precision (P) and the area under curve (AUC) of RF model are better in the ice flood risk assessment of four study areas. This presents that RF has significant advantages in solving the classification and processing problem of multi-dimensional ice flood hazard data. It can provide support for the analysis and evaluation of the situation of ice flood prevention and mitigation.