Theophile et al. (2026) Data Representation Shapes the Comparative Performance of XGBoost, Random Forest, and LSTM for Groundwater Head Prediction: A Case Study in Friuli Venezia Giulia, Italy

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2026
• Date: 2026-09-25
• Authors: Bimenyimana Theophile, Claudia Cherubini
• DOI: 10.3390/w18192390

Research Groups

[Information not available in the provided text.]

Short Summary

This study systematically compares three machine learning models (XGBoost, RF, LSTM-NN) for groundwater head prediction using a 26.5-year monitoring record, demonstrating that model performance is more dependent on input data representation and feature engineering than on the algorithm choice alone.

Objective

• To systematically compare the relative contributions of machine learning algorithms (XGBoost, RF, LSTM-NN) and input data representation for reliable groundwater head prediction.

Study Configuration

• Spatial Scale: Friuli Venezia Giulia Region, Italy (regional scale)
• Temporal Scale: 26.5 years

Methodology and Data

• Models used: Extreme Gradient Boosting (XGBoost), Random Forest (RF), Long Short-Term Memory Neural Networks (LSTM-NN)
• Data sources: Historical groundwater head monitoring records; derived features including climatic variables, seasonal encoding, polynomial transformations, differenced series, and climate-enhanced engineering.

Main Results

• Model performance depends strongly on data representation rather than algorithm choice alone.
• Under the autoregressive configuration (using only historical groundwater head), XGBoost achieved the highest accuracy (R² = 0.9998), outperforming RF (R² = 0.9960) and LSTM-NN (R² = 0.9097).
• When hydro-meteorological and engineered features were incorporated, LSTM-NN became the best-performing model (R² = 0.961, RMSE = 87.15, MAE = 59.75), followed by RF (R² = 0.9454) and XGBoost (R² = 0.9370).
• The climate-enhanced dataset consistently produced the highest accuracy by better representing delayed recharge and seasonal groundwater dynamics.

Contributions

• Systematic comparison of three widely used machine learning algorithms (XGBoost, RF, LSTM-NN) for groundwater head prediction.
• Demonstration that feature engineering and data representation are as critical as algorithm selection for achieving high model performance.
• Provision of practical guidance for developing reliable machine learning models for groundwater forecasting.

Funding

[Information not available in the provided text.]

Citation

@article{Theophile2026Data,
  author = {Theophile, Bimenyimana and Cherubini, Claudia},
  title = {Data Representation Shapes the Comparative Performance of XGBoost, Random Forest, and LSTM for Groundwater Head Prediction: A Case Study in Friuli Venezia Giulia, Italy},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18192390},
  url = {https://doi.org/10.3390/w18192390}
}