A machine learning approach to forecast CSO events with a random forest algorithm
Open Access
- 1 May 2026
- journal article
- Published by IWA Publishing in Journal of Hydroinformatics
- Vol. 28 (5) , 476-494
Abstract
The graphical abstract includes five boxes. The first one shows the input data used by the model, which are 5 min rain gage and water level sensor data. The second box shows the tree structure of the model and bars illustrating the importance of each input data. The third box shows that the predicted CSO occurrence can support proactive CSO management, predictive maintenance, and early warning systems. The fourth box relates the outcomes of the model to better CSO mitigation and planning. Finally, the fifth box summarizes the performance of the model in predicting CSO occurrence, which are F1-score = 0.91 - 0.92 and Accuracy = 0.93 – 0.95. Combined sewer overflows (CSOs) remain a major challenge for urban water management, with risks increasing under more frequent, intense rainfall. This study developed a random forest (RF) classification framework to forecast CSO occurrence in real time at two Quebec City outfalls (U051 and U12B). The model was trained on 5-min rainfall data from nearby rain gauges and water level data measurements at each outfall. Predictors were engineered to represent both short-term impact and antecedent conditions. The RF models achieved strong event-detection performance, with F1-scores of 0.92 (U051) and 0.91 (U12B) for the CSO occurrence class, with corresponding CSO-class precision of 0.94 and 0.91 and recall of 0.91 and 0.92, respectively. Feature importance results showed that lagged water level and recent rainfall intensity/accumulation metrics were the most influential predictors, with differences in dominant drivers between the two outfalls. Compared with traditional hydrodynamic modelling, the proposed RF approach provides a computationally lightweight alternative that does not require detailed network geometry or extensive calibration, making it easier to deploy and update in operational settings. Compared with deep learning approaches, RF offers faster training, robust performance with moderate datasets, and transparent interpretability through feature importance, enabling practical decision support for real-time CSO warning.Keywords
Funding Information
- Natural Sciences and Engineering Research Council of Canada (ALLRP 586388-23)
This publication has 39 references indexed in Scilit:
- Enhancing Rainfall-Runoff Simulation in Data-Poor Watersheds: Integrating Remote Sensing and Hybrid Decomposition for Hydrologic ModellingWater Resources Management, 2025
- Categorization of precipitation for predicting combined sewer overflows. Application to the City of MontréalJournal of Hydrology, 2024
- The Adoption of Random Forest (RF) and Support Vector Machine (SVM) with Cat Swarm Optimization (CSO) to Predict the Soil Liquefaction2024
- A state-of-the-art review for the prediction of overflow in urban sewer systemsJournal of Cleaner Production, 2023
- Forecasting and optimization for minimizing combined sewer overflows using Machine learning frameworks and its inversion techniquesJournal of Hydrology, 2023
- Costs and benefits of combined sewer overflow management strategies at the European scaleJournal of Environmental Management, 2022
- Uncertainty analysis of urban sewer system using spatial simulation of radar rainfall fields: New York City case studyStochastic Environmental Research and Risk Assessment, 2018
- Use long short-term memory to enhance Internet of Things for combined sewer overflow monitoringJournal of Hydrology, 2018
- Trees vs Neurons: Comparison between random forest and ANN for high-resolution prediction of building energy consumptionEnergy and Buildings, 2017
- Development of probability distributions for urban hydrologic model parameters and a Monte Carlo analysis of model sensitivityHydrological Processes, 2013