In order to realize intelligent quantitative assessments of urban fire risks and implement optimal allocation of fire resources, a random forest algorithm-based fire risk assessment method for urban buildings is proposed. The methodological framework is composed of three parts. Firstly, a fire risk score assessment model using the random forest algorithm was developed. The model is based on fire characteristic data of the Milesburg city in the United States, to numerically assess fire risks of buildings in the city. And then, fire risk levels of buildings were determined according to building types and the risk score, and relationships between the risk level and building properties were explored. In addition, a feature importance algorithm FI (feature importance) was designed to analyze fire risk factors and mine fire hazard characteristics with high impact factors, so as to guide the fire authority to enhance their work efficiency. Finally, interactive fire hotspot risk maps were designed and developed to facilitate the fire authority to quickly determine fire risks of an area or a building in the city, and accordingly improve fire inspection strategies and optimize allocation schemes of human and material resources.