In earthquake-prone regions, the rapid and accurate determination of damage levels in reinforced concrete (RC) buildings is crucial for emergency response and preventive measures following an earthquake. The ability to quickly assess the structural integrity of buildings not only aids in immediate post-disaster decision-making but also enhances long-term urban planning and resilience strategies. This study focuses on evaluating the damage classes (low, medium, high) of RC buildings using advanced machine learning methods. The evaluation is based on critical parameters outlined in the Turkish Regulations for Determining Regional Earthquake Risk Distribution of Buildings (6306), ensuring the assessment process aligns with international standards.
The parameters considered in this study include the number of stories, soil type, seismic impact, construction quality, and structural system type, specifically whether the building employs a frame system or a combination of frame and shear wall systems. Additionally, structural irregularities such as short columns, soft stories, plan irregularities, and heavy overhangs, along with topographic effects, are incorporated into the assessment. These factors are vital in understanding the vulnerabilities of RC buildings under seismic loads and are critical for accurate damage prediction.
This study will rigorously evaluate the damage classes of 250 reinforced concrete buildings using various machine learning algorithms. The selection of these methods is driven by the need to handle complex and multidimensional data while maintaining high accuracy rates. Random Forest employs an ensemble of decision trees to classify data, where the average of these trees helps to reduce errors. This method is particularly effective when there are complex relationships between variables. Support Vector Machines (SVM) create hyperplanes to classify data and aim to maximize the margin between these planes. SVM is known for its high performance in high-dimensional datasets and generally achieves high accuracy rates. Artificial Neural Networks (ANN) mimic the operation of neural cells in the human brain, utilizing a multilayered model. When combined with deep learning techniques, this method is highly capable of learning and generalizing complex data relationships.
These models will be trained on historical earthquake data, enabling them to predict the damage classes of new buildings with an expected accuracy exceeding 80%. The training process will be conducted iteratively on large datasets, continuously improving the model's performance.This machine learning-supported approach will enable the rapid identification and prioritization of high-risk buildings for detailed analysis, thereby facilitating more effective risk management.