Sundar et al. (2026) Combinatorial Analysis of Multi-Domain Feature Sets for Regional Monsoon Rainfall Prediction

Identification

• Journal: Lecture notes in networks and systems
• Year: 2026
• Date: 2026-01-01
• Authors: S. Sundar, M. Prathilothamai
• DOI: 10.1007/978-3-032-13003-7_26

Research Groups

• Department of Computer Science and Engineering, Amrita School of Computing, Amrita Vishwa Vidyapeetham, Coimbatore, India

Short Summary

This study empirically evaluates 2,047 feature group combinations to improve regional monsoon rainfall prediction using a hyperparameter-tuned XGBoost model. It demonstrates that combining meteorological features with oceanic and climatic inputs, such as lagged MJO and ENSO indices, significantly enhances prediction accuracy.

Objective

• To assess the impact of input feature composition on regional monsoon rainfall prediction accuracy by systematically evaluating various multi-domain feature set combinations.

Study Configuration

• Spatial Scale: Regional
• Temporal Scale: Monsoon season

Methodology and Data

• Models used: XGBoost (hyperparameter-tuned)
• Data sources: Curated sets of climate-related variables, including meteorological features, oceanic influences, broader climatic patterns, lagged climate indices (e.g., Madden-Julian Oscillation (MJO), El Niño-Southern Oscillation (ENSO)), cyclical temporal encodings, and event-based indicators.

Main Results

• Using only meteorological features, the model achieved an R² of 0.5801, a Root Mean Square Error (RMSE) of 10.8999 mm, and a Mean Absolute Error (MAE) of 4.9009 mm.
• When meteorological features were combined with oceanic and climatic inputs (particularly lagged MJO and ENSO indices and temporal signals), model performance significantly improved to an R² of 0.7606, an RMSE of 8.2297 mm, and an MAE of 3.8752 mm.
• The study evaluated 2,047 feature group combinations to identify optimal input compositions.

Contributions

• Provides an extensive empirical evaluation of the impact of multi-domain feature set composition on regional monsoon rainfall prediction accuracy.
• Demonstrates the significant improvement in prediction performance through domain-informed feature fusion, particularly with lagged MJO and ENSO indices and temporal signals.
• Offers a replicable approach for enhancing monsoon prediction models through systematic feature group design and empirical validation, addressing a gap in attention to input feature composition.

Funding

• Not specified in the provided text.

Citation

@article{Sundar2026Combinatorial,
  author = {Sundar, S. and Prathilothamai, M.},
  title = {Combinatorial Analysis of Multi-Domain Feature Sets for Regional Monsoon Rainfall Prediction},
  journal = {Lecture notes in networks and systems},
  year = {2026},
  doi = {10.1007/978-3-032-13003-7_26},
  url = {https://doi.org/10.1007/978-3-032-13003-7_26}
}