Prediction of Storm Water Overflow in Municipality Using Machine Learning*

In the past, storm water management relied on combined sewer systems that simultaneously collected both rainwater and wastewater. When these systems become overloaded, they discharge excess water into nearby bodies of waters, resulting in pollution. Although modern systems have largely mitigated this problem, many older pipes are still in use. Understanding when and where overflows occur is essential for effective mitigation and accurate reporting. Therefore, in this study, we apply machine and deep learning techniques, specifically logistic regression (LR) and Long Short-Term Memory (LSTM), to predict storm water overflow. The models were trained on IoT sensor data collected from the municipality of Horten, which was integrated with weather data. Our experimental results demonstrated that the LSTM model outperformed the LR model, achieving an accuracy of 87 %. Based on our results, we believe that our research could provide municipalities with timely and accurate insights, facilitating proactive measures to effectively address storm water management.