Improving of groundwater level estimation using SMOTE technique

Abstract

Accurate estimation of the groundwater level is very important for sustainable water
management and planning of water resources. Since the groundwater field studies are time
consuming and costly, the use of machine learning techniques for the groundwater level
estimation is proposed in this study. Also, synthetic minority oversampling technique
(SMOTE) algorithm was used to increase the success of the extra tree (ET) algorithm, which is
one of the machine learning techniques, in estimating the groundwater level. For this,
precipitation, elevation, slope, and curvature data, which are effective on groundwater, were
used. First, the groundwater level was classified as very low, low, medium high and very high.
When the result of the model developed with ET algorithm was evaluated, the accuracy value
was calculated as 0.53 and the Cohen's Kappa value as 0.36. Then, to increase the success of
the extra tree model, data irregularities were removed with the SMOTE algorithm. It was
observed that the accuracy of the SMOTE-ET model increased to 0.69 and the Cohen's Kappa
value to 0.62. When the results are evaluated, it is thought that the SMOTE algorithm increases
the success of the ET algorithm in estimating the groundwater level.