Evaluation of Data Mining Algorithms in Studying and Predicting the Qazvin Plain Aquifer Conditions

Due to population growth and agricultural development, predicting aquifer conditions and also determining factors effecting its performance are very important. In this research study, three algorithms of data mining package, i.e. IBM SPSS Modeler were applied for detecting models effective on predicting an aquifer conditions. Furthermore, human and natural factors were considered as input variables effective on predicting Qazvin plain aquifer conditions during 2001 to 2015. The accuracy of CHAID, CART, and CVM algorithms were 0.67, 0.75, and 0.74, respectively. The values obtained for different indices including TPR, TNR, ACC, FPR, FNR, FM, GM, and ER for CART decision tree model showed that the performance of the mentioned algorithm was better comparing to other algorithms. In addition, the results based on CART decision tree model show that the most important factors affecting water table fluctuations were air temperature, evapotranspiration and agricultural water demand. Thus, the developed model lead to predict water table depth in an aquifer and optimum management of the water resources.