Carbon footprint mapping of urban areas in Türkiye using hyperparameter-optimized machine learning techniques

This study predicted changes in carbon footprints at the city level in T & uuml;rkiye using the Bayesian method and hyper-optimized machine learning techniques with high accuracy and analyzed their spatial distribution. The dataset for 80 provinces in T & uuml;rkiye was searched. However, only the parameter data affecting the carbon concentrations in 2019-2022 and the carbon concentration amount dataset for 2019 could be accessed. The available dataset concerning parameters that affect carbon concentrations obtained from 80 cities in T & uuml;rkiye for 2019 was trained using hyper-optimized machine learning algorithms by the Bayesian technique (Ensemble Regression, Gaussian Process Regression, Gaussian Kernel Regression, Support Vector Machine Regression, Linear Regression, and Binary Decision Regression). Carbon footprint values were predicted for 2019, 2020, 2021, and 2022, and performance metrics were presented. In the application, while the ensemble regression technique obtained the lowest mean squared error value (2.31) for the training data, support vector machine regression obtained the lowest mean squared error value (13.8) in the test data. The Bayesian optimization algorithm significantly improved the success of all regression techniques applied in the study. After hyperparameter optimization, the regression methods' performance improved by 0.14 on average in terms of the coefficient of determination metric, 34.5 on average in the mean squared error metric, and 0.032 on average in the mean absolute relative error. The study will assist local governments in obtaining more accurate carbon emission estimates using less data and implementing climate action plans.