The accurate prediction of CO2 emissions from internal combustion engine vehicles is essential for reducing environmental impact and meeting strict emission regulations. This study introduces the Kolmogorov Arnold Network as a recent substitute to the Multilayer Perceptron algorithm for the predictive analysis of CO2 emissions from conventional vehicles. An exploratory data analysis was initially conducted, followed by a detailed description of the Kolmogorov-Arnold Network architecture and the selected model configuration. The study findings were then presented and analyzed in terms of performance and accuracy. This new approach provided improved performance in both prediction accuracy and computational efficiency, achieving a mean absolute error of 7.4250 on the train subset and 7.3404 on the test subset. The coefficient of determination was found to be 0.9823 for training and 0.9827 for testing. Furthermore, the performance metrics obtained by the suggested symbolic formula closely aligned with those obtained from the model.