An approach to rapid assessment of building seismic conditions utilizing supervised fuzzy C-means with consensus-based likert analysis

Two major earthquakes hit southeastern Türkiye on 6 February 2023. Damaged or collapsed buildings caused significant material losses, deaths, and hundreds of millions of tons of debris to be disposed of. Structures with a higher vulnerability to earthquake activity will suffer greater direct and indirect economic losses. It is essential to develop a methodology for rapid, cost-effective, and scalable seismic risk prediction of existing building stocks to determine which buildings are risky and require immediate detailed structural analysis, which is crucial to disaster risk reduction, economic planning, and building resilience. A novel approach combining Supervised Fuzzy C-Means (SFCM) Clustering with Consensus-Based Likert Analysis (CBLA) is presented to predict seismic risks using both structural indicators and expert evaluations. By aggregating expert ratings into a consensus-driven framework, CBLA enhances reliability by accounting for uncertainty and gradation in building conditions. A Likert scale was used to assess 200 buildings to train the model and validate the clusters, and 17 buildings were tested based on the most effective cluster centers. Cluster performance with two and four clusters was evaluated based on various initial centers, and then clustering validations were compared. Fuzzy_Rand performance in the two-class case scored 85%, whereas performance in the four-class case scored 74%. Results from the test data show that the model is very promising. This model is suitable for rapid, large-scale vulnerability screenings in urban and developing areas, reducing resource and time requirements. • Facades of existing buildings can be examined to classify their damage levels. • The C-means algorithm can classify existing buildings' damage levels. • Likert scales can be used to measure linguistic experts' opinions about damage levels. • Consensus follows human intuition when comparing Likert distributions.