Adaptive Network-Based Fuzzy Inference Systems Coupled with Genetic Algorithms for Predicting Soil Permeability Coefficient

This paper extends hybrid-type optimization models of genetic algorithm adaptive network-based fuzzy inference system (GA-ANFIS) for predicting the soil permeability coefficient (SPC) of different types of soil. In these models, GA optimizes parameters of a subtractive clustering technique that controls the structure of the ANFIS model’s fuzzy rule base. Simultaneously, a hybrid leaning algorithm is employed in the ANFIS, as a trained fuzzy inference system (FIS), which optimally determines the parameter sets of the examined FISs in ANFIS. Using an updated large database of SPCs consisting of 338 fine-grained, 178 mixed and 94 granular soil samples, GA-ANFIS framework constructs different models of predicting the permeability coefficient of respectively fine-grained, mixed and granular soils. A fuzzy C-mean technique has been used to cluster the entire data samples of each type of soil and divide them uniformly into training and testing data sets. Different prediction models of SPC have been trained and tested for each of the three soil types, and the appropriate models have been selected. The selected models have been compared with ANN and modified-by-GA empirical prediction models. Results show that the constructed GA-ANFIS models outperform the other models in terms of the prediction accuracy and the generalization capability.