Time Series Forecasting to Predict Pollutants of Air, Water and Noise Using Deep Learning Methods

Abstract

This paper analyzes the long-term air, water and noise pollution monitoring data using autoregressive integrated moving averages (ARIMA) modeling and an artificial neural network called long short-term memory network (LSTM). Different features of air, water and noise are predicted using these modeling techniques. This is known as time series forecasting. The accuracies of both the modeling techniques are compared with each other applied on the same datasets. We compare both the modeling techniques, and the models are evaluated with metric root-mean-squared error (RMSE) which is also called the objective function or loss function in ML lingo. The algorithm, equipped with prominent accuracy, is used to predict the future values of different features of air, water and noise. These predicted values are then compared with the health statistics, globally defined by the World Health Organization (WHO), to determine a fit and healthy life of an average human being. The harmful effects, possible disease threats and their precautions are also listed if the predicted values do not lie in the risk-free zone defined by the WHO. Most of the air, water and noise dataset is provided by the Central Pollution Control Board, Govt of India (CPCB) online. We will be using Python to execute this project.