Machine learning-based time series models for effective CO₂ emission prediction in India.

Area of Science:

• Environmental Science
• Data Science
• Climate Change Studies

Background:

• India is among the top global emitters of carbon dioxide (CO2).
• Per capita CO2 emissions in India are 1.80 metric tons, posing risks to health.
• Understanding and predicting CO2 emission trends is crucial for environmental policy.

Purpose of the Study:

• To analyze the detrimental effects of CO2 emissions in India.
• To predict India's CO2 emissions for the next 10 years.
• To compare the efficacy of various statistical, machine learning, and deep learning models for CO2 emission prediction.

Main Methods:

• Utilized univariate time-series data from 1980 to 2019.
• Employed statistical models: Autoregressive-Integrated Moving Average (ARIMA) and Seasonal Autoregressive-Integrated Moving Average with Exogenous Factors (SARIMAX), and Holt-Winters.
• Applied machine learning models: Linear Regression and Random Forest, alongside a deep learning Long Short-Term Memory (LSTM) model.

Main Results:

• LSTM, SARIMAX, and Holt-Winters models demonstrated the highest accuracy based on nine performance metrics.
• The Long Short-Term Memory (LSTM) model proved most effective for CO2 emission prediction.
• LSTM achieved a Mean Absolute Percentage Error (MAPE) of 3.101%, Root Mean Squared Error (RMSE) of 60.635, and Median Absolute Error (MedAE) of 28.898.

Conclusions:

• The Long Short-Term Memory (LSTM) deep learning model is highly recommended for accurate CO2 emission prediction in India.
• Comparative analysis confirmed LSTM's superior performance over other tested models.
• Accurate CO2 emission prediction is vital for developing effective climate change mitigation strategies.