Abstract
This study develops machine learning-based prediction models for carbon emissions in India, focusing on industrial applications and policy formulation. Accurate carbon emission forecasts are crucial for mitigating environmental impacts and guiding decision-making processes. Using historical data from the World Bank’s repository, time-series forecasting models were applied using Python to predict emissions across various sectors. The novelty of this study lies in integrating XGBoost, a powerful machine learning algorithm, with time-series forecasting to provide accurate and reliable predictions tailored to India’s industrial context. Seven critical transitions aligned with India’s low-carbon development strategies were identified, offering unique insights into emission patterns through trend analyses. Hyper parameters and evaluation metrics, such as the mean squared error, were optimized to enhance model performance. A comparative study with other machine learning models, including decision trees and random forests, demonstrated XGBoost’s superior accuracy. Furthermore, discussions on seasonality, trends, and stationarity of the data strengthened the forecasting model. While this study primarily focuses on industrial emissions, expanding the model to include non-industrial sectors such as agriculture and residential areas is recommended for comprehensive predictions. These findings are valuable for optimizing carbon management strategies, improving industrial efficiency, and informing mitigation policies. Future work will refine predictions using additional datasets and enhanced methodologies to improve long-term forecasting.