Analysis of ambient air quality data using machine learning algorithms and lstm architecture for two major traffic intersections of kolkata considering lichen as a bio-indicator

Abstract

In this study, an attempt was made to consolidate outcomes acquired from monitoring and mapping of lichen with quantitative analysis of AQI data acquired from continuous monitoring stations. Moreover, Index of Atmospheric Purity was evaluated utilising grid mapping method, in terms of assessment of frequency and cover of foliose and crustose lichens observed on various tree specimens. The wind direction and speed were also assessed in the study area that possessed major traffic intersections nearby. Based on seasonal variation for study period of 2019-2023, windrose analysis was conducted to track correlation among prevalent wind directions and lichen growth. Furthermore, air quality data acquired from CPCB’s continuous monitoring station at Jadavpur area were analysed using Machine Learning algorithms such as Linear Regression, Decision Tree Regressor, Random Forest Regressor, XGBoost Regressor and KNN Regressor. These models were compared on the basis of performance metrics R 2 , RMSE, MSE and MAE in order to suggest an effective model for predicting AQI. It was observed that Random Forest Regressor outperformed other algorithms for predicting AQI values effectively. Moreover, for time series forecasting of predicted data RNN-LSTM model has been utilised as it enables the acquire dependencies on several pollutants. Multivariate LSTM analysis was conducted on 7 significant pollutant concentrations to acquire suitable AQI prediction. Moreover, daily data for a span of 2019 - 2023 was estimated to acquire future forecasting of available data. Although LSTM model depicted higher accuracy, as compared to other ML algorithms the performance metrics were less. Therefore, with higher volume of data the accuracy metrics of the model could be enhanced.