OPTIMIZING FEATURE SELECTION FOR CLIMATE CARBON EMISSION INDICATORS: A MULTICOLLINEARITY-RESILIENT APPROACH WITH L1 AND L2 REGULARIZATION

Abstract

Carbon emissions, predominantly in the form of carbon dioxide CO₂, plays a vital role in driving climate change and its associated environmental consequences. This research go through into the origins, impacts, and potential remedies concerning carbon emissions, emphasizing their crucial role in global aim to combat climate change. In the context of Pakistan, carbon emissions primarily stem from energy production, industrial activities, transportation, and agricultural practices. This study specifically concentrates on the optimal selection of models for predicting carbon emissions in Pakistan. To achieve this objective statistical analyses perform based on various regression techniques. The response variable is taken as the consumption of CO₂ in solid, liquid, and gaseous forms, while predictors includes population, total value, and agricultural value added as a percentage of GDP, agricultural land area, urban population, gross fixed capital formation, industry value added, fertilizer consumption as a percentage of production, and GDP. The model selection process comprises five stages. In the initial stage, multicollinearity diagnosis is implemented through a correlation matrix. The second stage involves outlier identification using various statistical measures and graphical analyses, such as Box Plots. Subsequently, the third stage focuses on optimal model selection by using Ridge and LASSO regression analyses. The fourth stage. The efficient model selection based on criteria such as the sum of squared error (SSE), mean square error (MSE), root mean square error (RMSE), and forecasting efficiency assessed through the mean absolute percentage error (MAPE). The study result shows that on the basis of maximum  , minimum MSE, MAPE, LASSO regression is considered as the best technique according to the define selection criteria’s.