Modelling cable tray fires remains a complex issue, and existing models such as the FLASH-CAT model still have limitations due to the lack of information on certain input data. This paper discusses improvements in cable tray fire modelling by finding appropriate input data using an artificial intelligence (AI) technique. An AI-driven expert system was developed to assess missing data from a fire test database of 29 large scale cable tray fire experiments in open atmosphere. The expert system quantifies the dependencies between input and output data, helps identify influential parameters and refine model inputs. It also guides modelling efforts by identifying areas for model improvement. In this way, the applicability of an expert system in obtaining reliable input data for simulation tools to simulate a real fire scenario that has already been tested is demonstrated. Cable tray fire simulation results show improved accuracy compared with the use of default values for uncertain parameters.