Seismic Damage Prediction and Classification tool: Utilizing Adaptive Neuro-Fuzzy Inference System (ANFIS) and Geographic Information System (GIS)

Abstract

Seismic damage prediction of wide range of buildings is an important prioritization and classification tool for implementation of seismic risk and vulnerability assessment. This study has developed a rapid visual screening (RVS) tool using Adaptive Neuro-Fuzzy Inference System (ANFIS) model and Geographic Information System (GIS). Hospital and public-school buildings located in districts 5 and 6 have been selected for assessment. Technical calculation basis in the development of FEMA P-154 was followed in formulating the model using the neuro-fuzzy tool package in MATLAB. Six (6) building parameters and their resulting output (final score) were inputted for training. Sub-clustering FIS model showed the lowest RMSE and R2 result out of the five (5) FIS models compared. The model showed effective performance in classifying the buildings' damage grades having 0.2908, 0.8073 and 0.948 for RMSE, and for testing and overall coefficient of determination (R2) values, respectively. Performance metrics from the developed confusion matrix showed high range of values for sensitivity, specificity, precision, and accuracy, with maximum error rate and false positive rate of 0.11 and 0.25, respectively, in classifying the buildings' damage grade. These validate the model's performance and ability in classifying the damage grade of the buildings observed. The results were used to develop a digital mapping of the damage grade distribution of the selected buildings in the area using ArcGIS.