Deep Learning Approaches for Air Quality Forecasting: An LSTM and CNN Perspective

Nowadays, air pollution has become a significant and urgent environmental concern, because of human activities, growing industries, and urbanization. Prolonged exposure to elevated levels of air pollution might result in health issues. Hence, the ability to forecast pollution levels is essential for prompt action and efficient control of air quality. PM2.5 is a notable air pollutant comprised of minuscule particles, each measuring less than 2.5 micrometers in diameter. Rising PM2.5 concentrations exacerbate the risk to human wellness. This research examines the precise forecasting of long-term results of PM2.5 pollutants. The levels of PM2.5 help decrease air pollution and prevent its adverse effects. This study proposes a hybrid CNN-LSTM approach that uses the CNN algorithm to acquire spatial characteristics and the LSTM framework to derive temporal details. Also, this model analyzes data from five highly polluted stations in Delhi, and the data was collected from 01/01/2019 to 10/12/2020 from the Central Pollution Control Board in India. Correlation coefficients between PM2.5 and various substances aid in identifying comparable pollutants and evaluating their collective health effects. To assess the efficacy of these models, it is important to measure such as Mean Square Error and Mean Absolute Error (MAE). The hybrid model exceeds the baseline models, leading to the lowest error rate.