Evaluating statistical and machine learning models for river flow forecasting in Terengganu: a case study using facebook prophet, XGBoost, and random forest
Accurate forecasting of river flow is essential for effective flood management and sustainable water resource planning, particularly in regions influenced by seasonal monsoons like Terengganu, Malaysia. This study evaluates and compares the predictive performance of three forecasting models includin...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | en |
| Published: |
IEEE
2025
|
| Subjects: | |
| Online Access: | https://umpir.ump.edu.my/id/eprint/47305/1/Evaluating%20Statistical%20and%20Machine%20Learning%20Models%20for%20River%20Flow%20Forecasting%20in%20Terengganu%20A%20Case%20Study%20Using%20Facebook%20Prophet%20XGBoost%20and%20Random%20Forest.pdf https://umpir.ump.edu.my/id/eprint/47305/ https://doi.org/10.1109/AiDAS67696.2025.11212852 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate forecasting of river flow is essential for effective flood management and sustainable water resource planning, particularly in regions influenced by seasonal monsoons like Terengganu, Malaysia. This study evaluates and compares the predictive performance of three forecasting models including Extreme Gradient Boosting (XGBoost), Facebook Prophet, and Random Forest using monthly river flow data from the Dungun and Kemaman Rivers. The results reveal that the Random Forest model consistently outperforms the other two, achieving the lowest Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) at both river stations. Specifically, Random Forest achieved up to 41.5% lower RMSE and 17.1% lower MAE compared to XGBoost for Kemaman River. For Dungun River, Random Forest model outperforms the XGBoost model by 16.45% for RMSE and 16.17% for MAE. The results indicate that the ensemble-based nature of Random Forest provide greater robustness and accuracy in capturing peak-flow events and nonlinear dynamics of river flow series. These findings emphasize the effectiveness of machine learning models, especially Random Forest, in capturing the complex patterns of hydrological time series and highlight their potential integration into regional flood forecasting and water management systems. |
|---|