IoT-Based Flood Monitoring System Using Machine Learning Approach

This research aims to propose an effective IoT-based flood monitoring system using a machine learning approach (IM-FMS) for flood detection, monitoring and predicting. The proposed methods were demonstrated by using Arduino to collect real-time hydroclimatic data including ambient temperature, relative humidity (RH), Batu River’s water level and rainfall intensity, and then transmit wirelessly to Control Centre by Grove-Wio-E5 LoRa modules through LPWAN communication. Then, every eight seconds, Mosquitto MQTT connection transmits real-time data wirelessly to flood forecasting models, Node-RED IoT dashboard, and saves in local disc without an Internet connection. Node-RED provides a public-accessible and admin-accessible interface for monitoring current weather conditions and modifying subscribers. For data every three minutes, a Wi-Fi connection wirelessly transmits it to Google Spreadsheet for cloud storage. Telegram Bot and email are proposed to send flood alert notifications to respective subscribers. The system’s performance was evaluated in terms of machine learning (ML) models’ results and data transmission range via LoRa. Considering the whole dataset, we can observe that the random forest (RF) model’s accuracy (98.22%), precision (84.00%) and recall rate (70.00%) score are higher than the logistic regression (LR) model’s accuracy (92.74%), precision (29.09%) and recall rate (53.33%) score.