Evaluation of classification and decision trees in predicting daily precipitation occurrences

Due to the heterogeneous distribution of precipitation, predicting its occurrence is one of the primary and basic strategies to prevent possible disasters and their damages. Hence, this study aims at evaluating the capabilities of Logistic Model Tree (LMT), J48, Random Forest (RF), and PART classification algorithms in precipitation forecasts at Pars Abad station using previous 1-4 days data of meteorological variables. So, five scenarios were considered based on the cross-correlation function and partial autocorrelation function for validation of the studied methods in the period of 2004-2019. In general, by examining the Kappa, root mean squared error (RMSE), mean absolute error (MAE) indicators, scenario number 1 using the input parameters of 1-day lag was determined as the most appropriate scenario to predict daily precipitation. Also, the obtained results showed that the PART had better performance with more than 80% accuracy in precipitation forecasting. Moreover, the most accurate performance of PART was scenario 1 with Kappa = 0.2007, RMSE = 0.3879 and MAE = 0.2856. The conclusive results indicated that by implementing classification algorithms and decision trees and using meteorological data of the previous days, daily precipitation could be predicted accurately.