Evaluation of Linear Regression and Ridge Regression as Baselines for Predicting Indonesia's CO₂ Emissions

Abstract

Indonesia's annual CO₂ emissions have increased alongside economic growth and rising energy demand, making short-term prediction relevant for environmental planning. This study evaluates Linear Regression and Ridge Regression as transparent baseline models for predicting Indonesia's total annual CO₂ emissions using public data from Our World in Data for 1990–2022. The predictors are GDP per capita, primary energy consumption, and population. The data were split chronologically, with 1990–2015 as training data and 2016–2022 as test data. Ridge Regression used standardized predictors, and its alpha parameter was selected on the training period using TimeSeriesSplit. Model performance was evaluated using MAE, RMSE, and R², and was compared with Naive Forecast and Linear Trend baselines. Linear Regression achieved the best performance on the test data with MAE of 19.181, RMSE of 21.384, and R² of 0.887. Ridge Regression produced MAE of 27.693, RMSE of 33.758, and R² of 0.719, while still outperforming the simple baselines. These results indicate that although regularization is theoretically motivated under multicollinearity, it did not improve, and in this case reduced, out-of-sample accuracy on a small annual dataset. The findings should be interpreted as predictive performance rather than causal relationships.