Abstract : To establish a winter low-visibility classification forecast model for Guanghan Airport , this paper investigates factors influencing the formation and persistence of low-visibility events . Using the surface observation data from Guanghan Airport and the Wenjiang sounding data from 2017 to 2023 , this paper constructs a model by means of three machine learning algorithms , i . e . , Random Forest ( RF) , XGBoost and LightGBM . Prediction performance is improved through feature importance analysis and hyperparameter optimization , and sample balancing technique is used to reduce the bias of model products . The results indicate that :(1) Low-visibility frequency at Guanghan Airport peaks between 08 : 00 and 09 : 00 in winter , decreases notably during daytime and gradually increases after 18:00 , exhibiting distinct diurnal variation . (2) The feature analysis reveals that relative humidity and atmospheric stratification stability are the core impact factors for visibility. The 24 h changes in pressure and temperature affect visibility through dynamic processes , while mid-level humidity influences the persistence of low visibility via uplift effect. (3) After hyperparameter tuning , the RF model demonstrates optimal performance overall on the test set , with significantly superior classification accuracy , lower false negative rates and enhanced cross- dataset stability compared to other models . In conclusion , the Random Forest algorithm can effectively enhance the classification forecasting ability of low-visibility events , and feature engineering has revealed the impact pathways of local thermodynamic coupling mechanisms on visibility. This model can provide certain technical support for aviation weather warnings .