<p>This paper addresses the high costs and low processing efficiency associated with the mass production of compacted graphite iron (CGI). To solve these issues, machine learning techniques were employed to develop three cutting surface roughness prediction models: BP neural network, RBF neural network, and support vector regression (SVR). The research results indicate that the RBF neural network model had the lowest average error and the highest prediction accuracy, with a test accuracy of 0.9956 and a prediction accuracy of 0.9159. Based on this model, the influence of cutting parameters on surface roughness (Ra) was explored. The findings suggest that changes in the feed rate (<i>f</i>) significantly affect Ra. Within a certain range, increasing the cutting speed (<i>V</i>_<i>c</i>) can effectively mitigate the impact of increasing <i>f</i> on Ra, while decreasing the cutting depth (<i>a</i>_<i>p</i>) also reduces Ra, though to a lesser extent. These results provide valuable guidance for selecting and adjusting machining parameters, which is crucial for maintaining the high machining quality of CGI.</p>