The Effect of The Data Type on Anfis Results, Case Study Temperature and Relative Humidity

Abstract

In this study, the Adaptive Neuro-Fuzzy Inference System (ANFIS) was used to create models to predict mean relative humidity and temperature with the most suitable inputs. To find the most appropriate data type for these meteorological parameters both hourly-daily and raw-normalized data sets were used, and results were compared. The models were trained with 2014-2017 data observed at Kirkuk city station in Iraq, were checked with both 2018 data of Kirkuk and Sanliurfa city station in Turkey to investigate whether a model set with the data of a country could be used for another country data set. The execution of models was evaluated by using root mean square error (RMSE), mean absolute error (MAE), and determination coefficient R2. Among the two parameters, the temperature achieved the best performance using relative humidity and dew point as input variables. According to the results, daily normalized data had lower error values and higher R2 than hourly un-normalized data. Additionally, the results showed that the model performed successfully at the Sanliurfa city station on the temperature parameter because of similar climate conditions to Kirkuk city.