This study explores the extent to which data mining and statistical techniques might assist the Fire Service in detecting threshold and pattern changes in its spatio-temporal fire data. Three entirely different scenarios are investigated. A post-hoc search for patterns was made of fires of suspicious or unknown cause in an area where a subsequently convicted arsonist was known to be operating. The spatio-temporal occurrence of chimney fires was compared with local climate data looking for any threshold conditions which might trigger the seasonal changes in occurrence. Finally an attempt is made to measure the effectiveness of the Firewise programme, which involves fire fighters visiting schools to instruct students in fire safety. The before and after incidence of residential fires in proximity to schools visited is assessed to determine whether the programme has had any measurable effect. Different data mining techniques are applied to each scenario.

The literature on change-point detection is reviewed and the applicability of identified techniques to real time fire data is discussed. Software options are discussed. The results suggest that fire data, especially time and location data, would be adequate for the purposes of detecting change-points but the problem under investigation must be clearly defined. Interpreted data about the fire must be accurate and unambiguous if it is to be of assistance in identifying change-points.