Comparison of ANFIS and ANN for Estimation of Thermal Conductivity Coefficients of Construction Materials

Abstract

The determination of the thermal conductivity coefficient of construction materials is very important in terms of fulfilling the condition of comfort, durability of construction materials, the economy of country and individual. In this study, linear regression, adaptive neural based fuzzy inference system (ANFIS), and artificial neural networks (ANN) models were developed to estimate the thermal conductivity coefficient values fromthe surface density (dry specific gravity/thickness) and the unit weight of construction materials. Validations of the developed models were investigated by statistical analysis. In predictive models, while the lowest determination coefficient (R2) and the highest root mean square error (RMSE) were obtained from linear regression, the highest R2 and lowest RMSE were obtained from the ANFIS model. The results of the ANN model according to results of linear regression, while R2 increased by approximately 6%, RMSE decreased by 30-39%. The results of the ANFIS model revealed that while R2 increased by approximately 12%, RMSE decreased by 59-71%. As a result, it is suggested that together the surface density and unit weight with ANFIS, the most appropriate method in the used methods, can be used as an alternative approach to estimate the value of thermal conductivity.