Sign in Subscribe
Artificial neural networks

Development of Machine Learning and Deep Learning Models for Predicting Groundwater Levels in South Korea

Strategic Insights

— 2 min read

Research conducted at Duy Tan University in Da Nang, Vietnam, has led to the development of machine learning and deep learning models to predict groundwater levels in South Korea. The study utilized four machine learning and two deep learning algorithms to forecast groundwater levels in Bongseong well on Jeju Island, South Korea. The models were evaluated using five performance metrics, including root mean squared error (RMSE), correlation coefficient (CC), Nash-Sutcliffe efficiency (NSE), relative error (RE), and root relative squared error (RRSE). The results showed that the random forest model (RF3) performed the best in predicting groundwater levels, achieving an RMSE of 0.053 m, CC of 1.000, NSE of 1.000, RE of 1.114, and RRSE of 0.013.

Key Takeaways:

  • A new study has been conducted to develop machine learning and deep learning models for predicting groundwater levels in South Korea.
  • The research utilized four machine learning algorithms (SGB, RF, GRNN, and GMDH) and two deep learning algorithms (Deep ESN and LSTM) to forecast groundwater levels in Bongseong well on Jeju Island, South Korea.
  • The models were evaluated using five performance metrics (RMSE, CC, NSE, RE, and RRSE) to assess their predictive accuracy.
  • The random forest model (RF3) achieved the best predictive accuracy, with an RMSE of 0.053 m, CC of 1.000, NSE of 1.000, RE of 1.114, and RRSE of 0.013.
  • The study utilized the SHapley Additive exPlanations (SHAP) strategy and one-way Analysis of Variance (ANOVA) test to evaluate the predictive performances of the models.
  • The results of the ANOVA test revealed that the predicted values were extracted from the same population as the measured values based on all models in scenario 03.

Statistics:

  • Root mean squared error (RMSE) for the RF3 model: 0.053 m
  • Correlation coefficient (CC) for the RF3 model: 1.000
  • Nash-Sutcliffe efficiency (NSE) for the RF3 model: 1.000
  • Relative error (RE) for the RF3 model: 1.114
  • Root relative squared error (RRSE) for the RF3 model: 0.013
  • Five evaluation measures (RMSE, CC, NSE, RE, and RRSE) were used to assess the predictive accuracy of the developed models.

Sources:

  • Development of the machine learning and deep learning models with SHAP strategy for predicting groundwater levels in South Korea. Scientific Reports, 2025;15(1):35523.
  • Nature Portfolio, Heidelberger Platz 3, Berlin, 14197, Germany.
  • Duy Tan University, Institute of Research and Development, Da Nang, Vietnam.
  • Meysam Alizamir, Institute of Research and Development, Duy Tan University, Da Nang, Vietnam.
  • Sungwon Kim, Salim Heddam, Sun Woo Chang, Il-Moon Chung, Ozgur Kisi, and Christoph Kulls.

Read more