Comparison of Lazy Algorithms and M5 Model to Estimate Groundwater Level (Case Study: Plain Neyshabur)

In recent years and in many countries, overusing groundwater resources had been higher than their annual feeding amount. This issue caused drop in the groundwater levels, followed by drying wells, qanats and springs. In this study, given the importance of Neyshabur plain in supplying agricultural, industrial and drinkable water of the area, lazy algorithms of KNN, KSTAR and LWL and M5 tree model have been utilized under seven different scenarios in order to estimate groundwater level of this aquifer. To compare the results, the Statistical parameters of root mean square error, correlation coefficient and the average absolute error were analyzed. The results showed that the ‘f’ scenario which contains the volume of water discharged and total precipitation parameters is less efficient because the ground surface level parameter was not taken into account. In ‘a’, ‘b’ and ‘g’ scenarios, an optimum estimation has been maintained for the groundwater level by considering the parameters of total rainfall in the previous month, total rainfall in the last two months and the ground surface level. Among the models of lazy algorithms and M5 decision tree model, the ability of KNN model under ‘a’ scenario was more than other models in December (Azar) by the statistical parameters RZ=0/96 , RMSE= 6.56 and MAE= 3.53. Also, study of evaluation criteria showed that the LWL is not an appropriate model to predict the level of the water table.