Its Meteorology-Based Prediction Using LSTM Neural Network

Due to manmade activities and natural calamities, the air quality that we breathe is degrading. Air quality index (AQI) depends on multiple factors such as concentration of different pollutants in air, meteorological parameters, traffic density, etc. Using the Pearson Coefficient, it was established, that concentration of PM10, PM2.5 has perfect association with AQI having R values of 0.97 and 0.99, whereas concentration of other air pollutants, CO and O3 have R values of 0.51 and 0.42, respectively. Meteorological parameters like diurnal temperature, pressure have positive correlation with R values of 0.47 and 0.58. Other parameters, like humidity, dew point, and wind speed have a negative relationship with AQI, having R values of − 0.57, − 0.64, and − 0.43, respectively. Multiple combination of normalized and unnormalized input feature vectors, created using two different workflows, is used to implement LSTM models. It was observed that LSTM model with target variable and five meteorological parameters, diurnal temperature, pressure, humidity, dew point, and wind speed has outperformed all other combinations, achieving RMSE of 30.313 on scale of 0–500 and R2 score of 0.801. Furthermore, training RMSE and validation RMSE show that a larger number of epochs can help in improving model prediction accuracy.