A modified brain emotional learning model for earthquake magnitude and fear prediction

Brain emotional learning (BEL) model has been used frequently for predicting a quantity or modeling complex and nonlinear systems in recent years. In this research, two methods proposed for improving the efficiency of original BEL model using fuzzy rules, learning automata concepts and optimization algorithms. In the first proposed method, different optimization algorithms and continuous action-set learning automata (CALA) were used for finding the weights of BEL model, while in the second proposed model, the weights obtained using original rules of BEL model. In fact, in the second model finite action-set learning automata, CALA and different optimization algorithms were used for calibrating the learning parameters of the model. Also in the both proposed methods after extracting frequency features in thalamus, deep belief network is used in the sensory cortex for reducing the size of features. In addition, ANFIS is used for making fuzzy rules in the amygdala. The proposed models were used for magnitude and consequently fear prediction of the earthquakes. The results show that although both proposed methods are more accurate than original BEL model and could be used successfully, the second proposed model is more precise and reliable than the first proposed model.