Predicting India's CO2 Emissions from Vehicles in the Next 20 Years: A Comparative Study of Statistical and Deep Learning Models

Abstract

With the rising concerns over climate change and global warming, reducing carbon dioxide (CO2) emissions has become crucial, especially in rapidly industrializing and urbanizing countries like India. This study aims to predict India's harmful CO2 emissions for the next two decades using a combination of statistical and deep learning models based on univariate time series data. The research employs an Auto Regressive Integrated Moving Average (ARIMA) model alongside four deep learning models, including Recurrent Neural Network (RNN), Artificial Neural Network (ANN), BiLSTM-DNN, and ResNet models. Utilizing the Climate Analysis Indicators Tool (CAIT) dataset, covering India's CO2 emissions from 1990 to 2019, the study forecasts emissions from 2023 to 2042. The predictive models' accuracy is analyzed and compared using various metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R2 Score, and Mean Absolute Percentage Error (MAPE). The research identifies the most accurate model, which can serve as an effective tool for predicting India's CO2 emissions. The findings reveal that the ResNet and BiLSTM-DNN models outperform other models, exhibiting high R2 Scores and low MSE, MAPE, and MAE values, thereby holding promise for applications in the automotive industry and other sectors to mitigate air pollution and combat climate change.