IoT Based Air Pollution Monitoring & Prediction System
Mohammed Rakib, Sanaulla Haq, Md. Ismail Hossain, Tanzilur Rahman
Abstract
Air pollution has reached catastrophic levels in recent times and is increasing at an alarming rate. To address this problem, we propose a solution combining IoT and Machine Learning (ML) that will not only detect pollution levels in the atmosphere accurately but also predict future pollutant levels. The proposed solution consists of two parts: an IoT device with four sensors connected with a processing unit and a software model. The sensors can track the air quality around us by measuring Particulate Matter, Carbon monoxide, Ammonia, Temperature, and Humidity. A microcontroller processes the data from the properly calibrated sensors and sends them to realtime cloud storage through a Wi-Fi module. A remote server then fetches and analyzes the data. We then determine the prime pollutant from the data, which is Particulate Matter. Next, we use the Autoregressive Integrated Moving Average (ARIMA) model to predict the pollution levels from harmful gases & the Air Quality Index (AQI) of the next day with high accuracy. Precisely, we predict the 24-hourly observations of the following day after training our optimized model with 144 hourly observations of the previous six days. We then evaluate the model with MAPE (Mean Absolute Percentage Error) score, which is 2.82 percent for temperature, 4.70 percent for humidity, 6.92 percent for Particulate Matter 2.5, 10.12 percent for Carbon monoxide, 10.3 percent for Ammonia, and 5.79 percent for AQI. This implies that our model correctly predicted the values for all the parameters with an accuracy of 90 percent or more. We, therefore, believe that such a solution would be useful if a large-scale installation is done.