IoT-Based Real-Time Weather and Disaster Management System Using Machine Learning

Abstract

The article describes a basic IoT framework through the integration of ESP 323 samples, an array for several sensors, as well as an ensemble of machine learning algorithms for both environmental monitoring and disaster prediction. The IoT monitoring system incorporates MQ series gas sensors (MQ-135, MQ-2, MQ-7, MQ-9), ultrasonic distance sensors, and the OpenWeather API to monitor the environment for air quality (e.g. smoke detection), carbon monoxide levels and combustible gas levels, water level and seismic activity. One of the most significant advantages of this proposal is the machine learning pipeline, which consists of a total of ten separate classifiers and includes: Random Forest, Gradient Boosting, Support Vector Machines, and Neural Networks. The use of the ensemble models gives 94% or higher accuracy in predicting flooding events. Other features that are part of the proposed solution include: automated email alerting, real time visualization of sensor data via Web Dashboard, access to sensor data via ThingSpeak cloud service, and ability to analyze both historical and current data. The experimental testing indicates the ability of the proposed system to detect hazards using configurable alert thresholds, and 15 minutes of alert cool down period to avoid generating too many alerts (alert fatigue). Overall, the proposed architecture can provide a cost-effective, scalable solution to encompass both environmental monitoring and disaster early warning systems.