Abstract: Utilizing Meteorological data from Chengdu Shuangliu Airport (located 50 km from Tianfu Airport) and ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF), the low visibility forecasting model for Tianfu Airport is developed based on the random forest algorithm. The TrAdaBoost algorithm was used to improve the model, and the European fine grid forecast data was used to verify the effect. The research findings reveal that the model developed with the TrAdaBoost algorithm demonstrates a notable improvement of over 40% in the F2 score, achieving a score of 0.42, surpassing other methods. Key physical characteristic factors influencing the model performance include the M index, 6-hour pressure change, 2-meter dew point temperature difference, 700 hPa relative humidity, and 3-hour temperature change. Further analysis shows significant differences in the average values of these physical parameters between low visibility periods and the winter half-year (October-March of the next year), highlighting the distinctive characteristics of these factors during low-visibility weather and their statistical significance. More importantly, the model provides a solution for building forecasting models with small sample datasets and demonstrates promising applications in meteorological forecasting.