Evidential reasoning and machine learning-based framework for assessment and prediction of human error factors-induced fire incidents

More than 55% of recorded fire incidents in multi-unit residential buildings (MURBs) are induced by human error factors causing threats to the public and tremendous economic losses. This study developed a framework to quantitatively assess and predict human error factors that induce fire incidents (HEFs-FIs). The framework is based on four main steps: (i) identified the potential HEFs-FIs in MURBs and determined their relative frequencies; (ii) developed benchmarks to evaluate the HEFs-FIs in several cities; (iii) assessed the contribution of HEFs-FIs to the regional fire incidents using evidential reasoning; and (iv) developed artificial neural networks (ANN) and classification models to predict HEFs-FIs and fire origin. The developed framework is applied to seven cities of British Columbia, Canada, to show its applicability. Twenty-eight human error factors are found to induce fire incidents in MURBs. The most critical HEFs-FIs are determined for each city using the developed benchmarks. The evidential reasoning results discovered that 20.68% of the HEFs-FIs are very high-risk, including incendiary fire and smokers' materials. An ANN model is found to be very strongly correlated with a correlation coefficient of 82% in predicting HEFs-FIs and their origin location. Moreover, the ANN model is found to outperform the classifiers. The results help decision-makers take actions accordingly to enhance fire prevention and protection strategies. •Developed an integrated framework to investigate human errors-based fires.•Identified 28 potential human error factors that induce fire incidents.•Developed relative frequencies-based benchmarks for human error factors.•Determined 20.68% of human factors in British Columbia with a very high-grade risk.•Developed Neural Networks model with a correlation of 82% outperforming classifiers.