Abstract: 【Objective】This study aimed to develop a prediction model for occurrence of red spider mites on greenhouse cut roses， in order to provide technical support for early warning and precise control of red spider mite， and to offer a theoretical basis and practical reference for the intelligent management of pests and diseases in protected floriculture.【Method】Based on the environmental meteorological factors and field pest investigation data of greenhouse cut roses in Jinning District， Yunnan Province， from January 4， 2023， to January 7， 2024， the occurrence pattern of red spider mites and their key driving factors （internal temperature and relative humidity ranges） were analyzed. The resistance of 26 cut rose germplasms to red spider mites was evaluated and classified. Five machine learning models （K-nearest neighbors （KNN）， gradient boosting machine （GBM）， random forest （RF）， Bayesian generalized linear model （BGLM）， and support vector machine （SVM）） were employed to establish prediction models for red spider mite occurrence， and their prediction effects were evaluated.【Result】Red spider mite infestations of cut roses in Jinning District， Kunming exhibited a progressive pattern. In the late stage of infestation， leaves， stems， and buds were covered with white cobweb that impaired photosynthesis and led to susceptible abscission of leaves and buds. The red spider mite occurrence was seasonal， with high incidence in winter and summer， especially in January and July； and low incidence was observed in spring and autumn， with the lightest in April and May. An environment with a relative humidity of 55.5%-65.5% and a greenhouse internal temperature of 22-24 ℃ was conducive to the outbreak of red spider mites. According to the harm index， the 26 varieties could be classified into five resistance clusters， among which Firebird， Luoshen， The Pride， Miss Julia， and Cynthia belonged to the high-resistance （HR） group. For predicting the highly susceptible （HS） resistance group， the RF and GBM models demonstrated the best performance， with both the coefficient of determination （R²） and model efficiency index （EF） above 0.80. Their root mean square error （RMSE） ranged from 3.56 to 3.57， normalized root mean square error （NRMSE） from 64.28 to 66.84， and mean absolute error （MAE） from 2.08 to 2.15.【Conclusion】RF and GBM models show superior prediction accuracy for red spider mite in greenhouse cut roses compared to the other three models. An ensemble model integrating RF and GBM can be utilized in an early warning system for rose pests and diseases. By embedding this system into a smart agricultural platform， scientific pest management can be achieved， providing decision-making support for precise pest control in rose cultivation areas across different climatic zones.