A Deep Learning Based Model to Predict PM Concentration

Particulate matter PM10 is the most effective air pollutant and has a serious impact on health and the environment. Hence, it is important to predict the level of PM10 concentration for the protection of the population. This study employs deep learning models such as MLP, GRU, and LSTM for the prediction of PM10 concentration at different time intervals. The main input variables for the proposed models are 15mins, 30mins, 1hr, 4hr, 8hr, 24hr measurements of PM10 concentration. The time series data is collected from the Central Pollution Control Board (CPCB), New Delhi, for five years, from 2016 to 2021. We use different performance metrics, such as the coefficient of determination (R2), normalized root mean square error (nRMSE), mean absolute percentage error (MAPE), to compare how well different deep learning techniques work. The LSTM model achieved higher predictive accuracy than the GRU and MLP models in estimating PM10 concentrations at different time intervals with R2 value ranging from 0.575 to 0.963, as shown by the experiments. Other models only achieved predictive accuracy of 0.575 to 0.963. In this study, a promising and cost-effective method for accurately predicting PM10 concentrations was developed and proposed.