The work aims to clarify the impact of temperature and humidity on the moisture change in loosening and conditioning process. K-means clustering analysis was used to divide the environmental temperature and humidity range in Zhangjiakou City and gradient enhancement decision tree was used to select parameters related to the moisture change in loosening and conditioning as characteristic variables. The prediction model based on machine learning was constructed for the moisture change in loosening and conditioning under different temperature and humidity ranges. Simultaneously, the impact of workshop temperature and humidity on moisture change was explored and the final model was optimized. According to the temperature and humidity of the external environment, it could be divided into four ranges throughout the year, including low temperature and medium humidity, medium temperature and low humidity, high temperature and high humidity, and medium temperature and medium humidity. The difference of moisture change in loosening and conditioning between the medium temperature and low humidity range and the high temperature and high humidity range was the most significant. Least Squares Regression, Support Vector Regression, Decision Tree and Random Forest prediction models were constructed for the moisture change in loosening and conditioning in the ranges of medium temperature, low humidity, and high temperature, high humidity. Through comprehensive comparison, it was found that the SVR and RF prediction models had the highest accuracy. After adoption of workshop temperature and humidity as the characteristic variable, the prediction accuracy of the SVR and RF models for the moisture change in loosening and conditioning increased, with R2´ (difference between R2 and 1) respectively decreasing by 25% and 46%, indicating that the temperature and humidity in the workshop had significant impact on the moisture change in loosening and conditioning process. After optimization of prediction models of the SVR and RF for moisture change in the medium temperature and low humidity range and high temperature and high humidity range, the R2´ of final model respectively reached 0.08 and 0.04. The moisture change in loosening and conditioning process is affected by external and workshop temperature and humidity. The RF model for medium temperature and low humidity range and SVR model for high temperature and high humidity range constructed based on different temperature and humidity have good fitting effects and accurate prediction on moisture change and can be reliably applied to predict the moisture change in loosening and conditioning. In addition, it can be expanded to be used in the moisture prediction and control of paper cigarette packaging materials, which has important theoretical significance and practical application value for improving the quality of rolled and packed tobacco.