Comparison of rapid evaluation methods for residential buildings in Bingöl province by using artificial neural networks

The world population has increased by 3 billion people since the baby born as the 5 billionth baby in Zagreb in July 1987. The birth of the 8 billionth baby was announced by United Nations in 2022. The incredible rate of increase in the population of the world, inevitably, causes an increase in residential areas that are often prone to natural disasters, including earthquakes. As a result, despite the significant advances in natural disaster modelling and risk mitigation, the loss of life and property in disasters continues to increase. The most important difference brought by the new policy determined with the Yokohama Conference held in 1994 is "Preparing National Action Plans for Risk Identification and Risk Mitigation". Seismic risk assessment is a critical link in reducing the loss of life and property from earthquakes. The need to determine this relationship has led to a rapid increase in demand for useful, reliable, and flexible digital tools and software. In this study, an artificial intelligence-based and open-source estimation tool developed for seismic hazard and risk assessment for Bingöl province of Turkey. This software can calculate loss and damage distribution for a deterministic earthquake scenario. Estimation of seismic risk mainly depends on three components: seismic hazard, exposure, and vulnerability. The second component is of particular importance as it may be possible to directly reduce the seismic risk by intervening with appropriate reinforcement solutions. In this study, a new vulnerability estimation tool was developed for the reinforced concrete building stock in Bingöl by using statistical information collected from real buildings that experienced the 2003 Bingöl Earthquake. A machine learning approach based on Artificial Neural Networks has been adopted, and the risk of possible earthquakes has been investigated based on some negativity parameters that have been determined to cause damage in the light of previous earthquake experiences for damage status criteria. When the results were examined, it was understood that the trained model made predictions with high accuracy. This was found promising, at least for the classification of the existing building stock, which is very high, before the earthquake.