Dry machining of AISI H13 steel is widely used in industrial processes due to its thermal and mechanical resistance. However, the absence of cutting fluids directly impacts surface roughness ( \(\varvec{R}_{\varvec{a}}\) ), making it crucial to predict this variable to optimize the final quality of machined parts. This study aimed to compare the performance of machine learning (ML) and GPT models in predicting \(\varvec{R}_{\varvec{a}}\) during dry turning of AISI H13 steel. The methodology involved applying an experimental design with cutting parameters (speed, feed rate, and depth of cut) and measuring \(\varvec{R}_{\varvec{a}}\) under controlled conditions. ML models (deep neural network - DNN and decision tree - DT) and three GPT variations (GPT-4 o1, GPT-4 o1mini, and GPT-4o) were trained and evaluated using metrics such as MSE, RMSE, and \(\varvec{R}^{\varvec{2}}\) . The results revealed that ML models, particularly DT, outperformed the GPT models across all metrics. DT demonstrated the lowest MSE and highest \(\varvec{R}^{\varvec{2}}\) , showcasing its superior ability to capture nonlinear patterns robustly and accurately. The DNN also performed well but was limited by the small dataset, which constrained its generalization capabilities. Among the GPT models, GPT-4 o1 was the most accurate but remained less effective than the ML models. In conclusion, DT proved to be the most suitable model for predicting \(\varvec{R}_{\varvec{a}}\) in dry machining, while adjustments to GPT models could enhance their performance.
