Using feature selection algorithm for prediction of evapotranspiration with the lowest data

In the present study the most effective climatological parameters for prediction of evapotranspiration using feature selection algorithm in Darab city located in southwest of Fars province was selected. In the first stage, the values of evapotranspiration were calculated based on FAO Penman-Montith method, then using feature selection method, the most effective parameters were selected among all effective parameters to evapotranspiration prediction based on FAO Penman-Montith method. Using Best First, Greedy Stepwise and Ranker as the most famous methods of feature selection the most effective parameters from 120 data were selected. Also in order to investigate of the error of each method to choose the best method Naïve Bayes, J48 and LMT was used. Using minimum, maximum and average temperatures, relative humidity, sunshine and maximum sunshine hours, wind speed, clear sky solar radiation (Rso), (75% of solar radiation in the upper atmosphere), the evapotranspiration was predicted. The results show that Ranker method with Relifef- Attribute-Eval in Naïve Bayes, J48 and LMT method had the lowest error. So maximum sunshine hours, maximum and average temperatures were found to be the most effective parameters for prediction of evapotranspiration. Using feature selection algorithm can be useful to predict of evapotranspiration in regions with limited data and save time and money.