Integrated ANN-cross-validation and AHP-TOPSIS model to improve earthquake risk assessment

The current study presents a novel combination of artificial neural network cross-validation (fourfold ANN-CV) with a hybrid analytic hierarchy process-Technique for Order of Preference by Similarity to Ideal Solution (AHPTOPSIS) method to improve the earthquake risk assessment (ERA) and applied it to Aceh, Indonesia, to test the model. Recent studies have suggested that neural networks improve probability mapping in a city scale. The network architecture design with probability index remains unexplored in earthquake-based probability studies. This study explored and specified the major indicators needed to improve the predictive accuracy in probability mapping. First, probability mapping was conducted and used for hazard assessment in the next step. Second, a vulnerability map was created based on social and structural factors. Finally, hazard and vulnerability indices were multiplied to produce the ERA, and the population and areas under risk were calculated. Results show that the proposed model achieves 85.4% accuracy, and its consistency ratio is 0.06. Risk varies from very high to high in the city center, approximately covering an area of 23% (14.82 km(2)) and a total population of 54,695. The model's performance changes on the basis of the input parameters, indicating the selection and importance of input layers on network architecture selection. The proposed model is found to generalize better results than traditional and some existing probabilistic models. The proposed model is simple and transferable to other regions by localizing the input parameters that contribute to earthquake risk mitigation and prevention planning.