Predicting Carbon Emissions in Supply Chain Operations Using Regression Forecast Emissions Through Machine Learning

Abstract

Climate change and increasing carbon emissions have become one of the most important issues in the world and the operations of the supply chain, especially transportation, manufacturing, warehousing, packaging and energy consumption are among the significant sources of greenhouse gas emissions. With the growing complexity, globalization and data-intensive nature of supply chains, the previous method of estimating emissions has not served as a sufficient method of predicting future carbon emissions. This MBA project fills this gap by offering a machine-learning-based regression forecasting system to seek carbon emission in supply chain activities and thus aid in informing data-driven sustainability decision-making. The main aim of the proposed study is to create and test machine learning regression models that can be used to predict carbon dioxide (CO2) emissions using main supply chain functions effectively. The study is aimed at discovering the most important operation drivers of emissions, evaluating the predictive power of machine learning models compared to the deterministic models and showing how predictive analytics can aid in sustainable logistics planning and environmental regulations. The proposed study is based on the general framework of the global sustainability programs, ISO 14001:2015 environmental management standards and the changing climate and environment policies in Pakistan especially in the Punjab. The study has a quantitative and predictive approach methodologically with a hybrid modeling framework comprising of regression analysis and machine learning and time-series predictive methods. The data of the carbon emissions are further broken into the long run trend and the short-run residual data. Structural pattern of emission is captured using parametric and non-parametric regression models whereas short-term fluctuations are captured by using linear and nonlinear time-series models, such as autoregressive and neural-network-based models. Ensemble machine learning models like the Random Forest are introduced in order to deal with the nonlinearity and enhance predictability. The analysis is performed with Python and the performance was measured with the standard error indicators like MAE, RMSE and MAPE.