Forecasting Underground Water Levels: LSTM Based Model Outperforms GRU and Decision Tree Based Models

Underground water is too important in human life from various perspectives, but water depletion is a significant problem worldwide. The underground water level is decreasing daily and often increasing in some places. In order to save water, it is essential to observe underground water levels, so we proposed a methodology based on machine learning and deep learning to forecast underground water levels. This study used machine learning algorithms, including Random Forest and XGBoost Regressor, and deep learning algorithms like LSTM and GRU. We also looked into the effects of such algorithms on time series data. In terms of Mean absolute error (MAE) and Root Mean Squared Error (RMSE) score, the LSTM-based model out-performed the GRU-based model, Random Forest Regressor, and XGBoost Regressor. In our study, deep learning algorithm-based models outperformed traditional machine learning algorithms on time series data of water level. Specifically, LSTM outperformed other models in estimating groundwater levels.