This paper focuses on fire prevention in UCG processes within active mines by means of temperature forecasting by a multivariate adaptive regression splines (MARS) approach. The main aim was to develop a model to forecast the temperature of the syngas with one hour of anticipation based on information from different parameters measured every hour (snapshots) during the experiment. As the response time of the syngas temperature to modifications in the composition/amounts of the gasifying agent is very short, this will reduce the temperature if necessary while keeping it as high as possible within the safety parameters, as UCG is a strongly exothermic process. The same model can be used to prevent undesired drops in the temperature of the syngas, as low temperatures could increase the precipitation of contaminants, causing a slowdown in the syngas flow and thus decreasing its calorific value. Temperature forecasting was achieved successfully; thus, the use of artificial intelligence and, specifically, a supervised learning predictive technique such as MARS models will allow active mines for the first time to adequately prevent the risk of fires while obtaining the best syngas quality in UCG processes. (C) 2018 Elsevier Ltd. All rights reserved.