عنوان مقاله [English]
Due to the location of Iran in dry and semi-arid climate, heterogeneous distribution of precipitation and also the occurrence of a climate change phenomenon has caused phenomena such as floods, drought, desertification and dust production and also creating the different economic, social and environmental damages. One of the primary strategies to reduce these losses, is prediction of the precipitation events. The goal of the present study is monthly precipitation prediction with using data mining methods of decision tree (M5) and K-Nearest Neighbor (KNN) algorithms and Comparing these methods in order to determining more efficient method in the field of predicting the precipitation using monthly meteorological data of Birjand synoptic station during the statistical period 1961-2010 in three cases the raw data, the three-year moving average and the five-year moving average in the Weka software. The results showed that in all defined scenarios, the tree model M5 has more ability than the KNN model to predict the monthly precipitation of the station. Also after investigation of the evaluation criteria R, RMSE, MAE and NS, the fifteenth scenario with input variables such as mean difference of maximum and minimum temperature, average relative humidity, average wind speed and cooling degree days (base 21 ° C) in every month was determined as the best scenario for predicting the same month precipitation. Also the obtained results from comparing the defined scenarios in each model in three states raw data, three-year moving average and five-year moving average show that in most scenarios The five-year moving average on average, with the values of R=0.90445, RMSE=6.0543 and MAE=4.78035 in the M5 model and on average, with the values of R=0.83689, RMSE=7.69825 and MAE=5.595 in the KNN model offers more accurate prediction of monthly Precipitation.