Using ANFIS for selection of more relevant parameters to predict dew point temperature

In this research work, for the first time, the adaptive neuro fuzzy inference system (ANFIS) is employed to propose an approach for identifying the most significant parameters for prediction of daily dew point temperature (Tdew). The ANFIS process for variable selection is implemented, which includes a number of ways to recognize the parameters offering favorable predictions. According to the physical factors influencing the dew formation, 8 variables of daily minimum, maximum and average air temperatures (Tmin, Tmax and Tavg), relative humidity (Rh), atmospheric pressure (P), water vapor pressure (VP), sunshine hour (n) and horizontal global solar radiation (H) are considered to investigate their effects on Tdew. The used data include 7 years daily measured data of two Iranian cities located in the central and south central parts of the country. The results indicate that despite climate difference between the considered case studies, for both stations, VP is the most influential variable while Rh is the least relevant element. Furthermore, the combination of Tmin and VP is recognized as the most influential set to predict Tdew. The conducted examinations show that there is a remarkable difference between the errors achieved for most and less relevant input parameters, which highlights the importance of appropriate selection of input parameters. The use of more than two inputs may not be advisable and appropriate; thus, considering the most relevant combination of 2 parameters would be more suitable to achieve higher accuracy and lower complexity in predictions. In the final step, comparisons between the predictions of the ANFIS model using the selected inputs and other soft computing techniques demonstrate that ANFIS has a higher accuracy to predict daily dew point temperature.