Groundwater Level Prediction Using Support Vector Machine and M5 Model Tree-A Case Study

Abstract

Groundwater is a vital source of fresh water, supporting the livelihood of over two billion people worldwide. Rapid population increase in urban areas has increased the demand for groundwater resources. This study aims to assess the groundwater level in a rapidly growing urban area, Kazhakkoottam in Trivandrum district. 80% of the total population in this area depends on groundwater for their daily needs. This study aims to develop two models: M5 Model Tree and Support Vector Machine (SVM) for the simulation of groundwater level (GWL) in the study area. The parameters considered in the development of the model are rainfall, temperature, humidity and evapotranspiration. In addition to these hydrological parameters population data was also considered for identifying the relationship between the city growth and groundwater level fluctuations. The fifteen years data were used for training the developed model and six-year data for validating the developed model. The best value of RBF kernel C and γ obtained for SVM was 12 and 4.5 respectively. The statistical parameters (Root Mean Square Error (RMSE), Mean Absolute error (MAE), and coefficient of determination (R2)) used for validation were found to be within acceptable limit for both the models. It is found that SVM performs better in simulating groundwater level compared to M5 Model Tree based on statistical parameters estimated. The time series prediction of future groundwater levels up to 2025 was also done using SVM. The predicted values indicates that the groundwater level is declining with the growing population.

Keywords: Groundwater level, Support vector Machine, M5 Model Tree, Time series prediction