Knowledge on fire ignition causes and their spatiotemporal patterns can greatly enhance the efficiency of fire management and fire strategies. In France, the majority of forest fire research is based on a 2x2 km gridded database that provides amongst other information, the cause of fire ignition. According to the same database however, approximately 75% of all fires between 1973 and 2020 were recorded without a cause of ignition. Therefore, information on fire causes for a very large part of the fires that were recorded in the last 50 years is not taken into consideration and can potentially provide significant evidence on patterns of different fire ignition causes. In the current study, we used for the first time a point geodatabase in order to predict the cause of fire ignition by applying several machine learning methodologies and modelling multiple environmental and anthropogenic drivers. As arson fires are of particular interest in SE France since they are the most frequent and cause the largest volume of burned area, we plan to analyse their spatiotemporal evolution over the last decades.