In this research development and analysis of Long Short-Term Memory (LSTM) neural networks for predicting the concentrations of air pollutants PM2.5 and NO2 in Tuzla based on meteorological parameters and previous concentrations of pollutants were done. The dataset covers heating season for 2022 with two time periods, the first from 01 January 2022 until 16 April 2022 and the second from 15 October 2022 until 31 December 2022. Input variables were temperature, air pressure, humidity, and pollutant concentrations PM2.5 and NO2 from the previous 3 h. Performances of developed models were evaluated using coefficient of determination (R2), mean squared error (MSE), and mean absolute error (MAE). SHAP (SHapley Additive exPlanations) analysis was used to assess the impact of input variables for each pollutant. The results showed that there were complex relations between input and output variables for both time periods. One-way ANOVA and Games-Howell tests were conducted to assess monthly variations in concentrations of air pollutants PM2.5 and NO2 and to test whether there were differences in means of PM2.5 and NO2 concentrations for each month. The research findings showed that there were statistically significant differences in concentration of pollutants PM2.5 and NO2 across months emphasizing seasonal variations. Developed LSTM models showed high performances with values of R2 74.46% and 78.24% for the first and the second time periods respectively. It was shown that LSTM neural networks are adequate tool to describe complex relations between concentrations of air pollutants and meteorological parameters.