Abstract: To address the problems of unscientific setting of hot air temperature during the unstable state stage, large fluctuations in outlet tobacco temperature, and long duration in the loosening and conditioning process, the relationship between hot air temperature and outlet tobacco temperature were modeled and optimized. Based on 170 sets of test data, linear regression and random forest algorithms were employed to quantify the temperature deviation in the unstable state stage by combining control surface error. A prediction model was established and the optimal hot air setting value at the unstable state stage was deduced, providing the basis for optimizing the temperature control of outlet tobacco leaves. The results showed that hot air temperature exhibited a significant positive correlation with the outlet tobacco temperature. The increase of the hot air set value accelerated the outlet temperature response and shortened the unstable state duration. The random forest model outperformed the linear model in fitting performance, achieving high prediction accuracy with a mean error less than 0.2 ℃. The proposed "prediction & optimization" strategy contributed to improving temperature control accuracy and stability in the conditioning process, providing a theoretical and practical foundation for digital control and intelligent adjustment in cigarette primary processing.