Deep Learning-Based Air Pollution Forecasting System Using Multivariate LSTM

Abstract

Air quality forecasting is an important way to safeguard public health since it provides ample notice of harmful air pollutants in a given location. Exhaust emissions from gas-guzzling vehicles, industrial pollution, and wildfires obfuscate some of the country's most dramatic vistas and can pose a significant hazard. If forecasts are trustworthy and accurate enough, they can be a crucial component of an air quality control system that complements more conventional emissions-based strategies. This study uses multivariate LSTM Time series forecasting techniques to understand and apply multiple variables together to contribute more accuracy to forecasting due to its proven track record of success with time-series data. The factors that are found to have a major effect on the air quality are dew point, wind speed, temperature, snowfall and rain conditions. In particular, this model focuses on predicting and forecasting the concentrations of PM2.5 (Atmospheric Particulate Matter with a diameter of less than 2.5mm) as this pollutant has a major effect on human health worldwide.