Fog occurrence is a common phenomenon during the winter season in Sofia airport (alt. 531 m). Therefore, fog forecasting up to 3 h lead time are urgently needed in order to maintain air transport. Stochastic models with 0.5, 1, 2 and 3 h lead time for fog forecast occurrence (less than 1 km horizontal visibility) at Sofia airport based on binary logistic time series regression analysis are developed and discussed. The probabilities of fog occurrence versus non occurrence are estimated using historical meteorological observations. Data with different time scale for the period 01.01.2007–31.08.2023 were used for this purpose such as: a) half-hourly aeronautic observations for fog, temperature, dew point, wind speed and pressure from METAR reports at Sofia airport; b) 3-hourly observations from the SYNOP data for temperature, dew point, wind speed, pressure and fog measured at the Central Meteorological Station (CMS) (alt. 595 m) of NIMH, Sofia; c) aerological sounding data for temperature, relative humidity, wind direction and speed at standard pressure levels 925, 850 and 700 hPa measured at 12:00 GMT at CMS of NIMH, Sofia. Various lags from 1 to 6 of these meteorological data serve as input predictors in the models. The selection of most significant model predictors is based on lasso penalization. The generalized cross-validation technique is used in order to evaluate the model performance over testing data sets. Standard measures widely used in atmospheric sciences and statistical learning for forecast model assessment are used. The developed models explain a high percentage of the fog data variation and are reliable tools for fog forecasting.