In this study, the levels of sulfur dioxide (SO2) and nitrogen dioxide (NO2) were measured indoors and outdoors using passive samplers in Tymar village (20 homes), an industrial area, and Haji Wsu (15 homes), a non-industrial region, in the summer and the winter seasons. In comparison to Haji Wsu village, the results showed that Tymar village had higher and more significant mean SO2 and NO2 concentrations indoors and outdoors throughout both the summer and winter seasons. The mean outdoor concentration of SO2 was the highest in summer, while the mean indoor NO2 concentration was the highest in winter in both areas. The ratio of NO2 indoors to outdoors was larger than one throughout the winter at both sites. Additionally, the performance of machine learning (ML) approaches: multiple linear regression (MLR), artificial neural network (ANN), and random forest (RF) were compared in predicting indoor SO2 concentrations in both the industrial and non-industrial areas. Factor analysis (FA) was conducted on different indoor and outdoor meteorological and air quality parameters, and the resulting factors were employed as inputs to train the models. Cross-validation was applied to ensure reliable and robust model evaluation. RF showed the best predictive ability in the prediction of indoor SO2 for the training set (RMSE = 2.108, MAE = 1.780, and R2 = 0.956) and for the unseen test set (RMSE = 4.469, MAE = 3.728, and R2 = 0.779) values compared to other studied models. As a result, it was observed that the RF model could successfully approach the nonlinear relationship between indoor SO2 and input parameters and provide valuable insights to reduce exposure to this harmful pollutant.