Application of Machine Learning Methods in the Analysis of Building Emergency Evacuation at Early Design Stages

Building Evacuation Assessment is one of the fields of research that has been of controversial fields of research in recent years. Along with common tools for building evacuation modelling, new techniques have also been adopted recently in this field. In this research, a new method based on the application of machine learning algorithms has been adapted for the prediction of the evacuation time. Also, a machine-learning-based analysis has been conducted to assess the effect of each architectural factor on the RSET (Required Safe Egress Time), using the schematic architectural layouts. The applied methods include XGBoost model, Gini importance analysis and Shapley Additive Explanations. The generated predictive model can predict the RSET of the models, with the mean absolute error of 30 seconds and R2 value of 0.94. The main influential factors on the models' outputs are the areas of high-density occupancies with occupant load factors of 0.5, 0.7 and 1.4, corridors areas and the main egress widths of the building floors. The proposed method is a reliable alternative for time-consuming building evacuation simulations.

Application of Machine Learning Methods in the Analysis of Building Emergency Evacuation at Early Design Stages Keywords:

Building Evacuation , Fire Emergency , Explainable Artificial Intelligence , XGBoost , Shapley Additive Explanations