Maintaining air quality is essential to life on Earth and is critical to public health protection. Based on that, forecasting air quality is a commonly handled urgent research area. Long Short-Term Memory (LSTM) is commonly used for air quality forecasting, where the utilized data is in the form of a time series. Nevertheless, existing studies have not thoroughly addressed missing data, which is an inevitable problem, particularly in real-life data sets. To strive to make up for this lack, we aimed to forecast PM2.5 air pollutant in Eskisehir using LSTM and different missing data imputation methods. Different imputation methods are considered to fill missing observations with plausible values, namely Backward Fill, Forward Fill, Zero Fill, Median Fill, Mean Fill, Random Fill, Interpolation, and Kalman Filtering. Our results indicate that the best forecasting performance is achieved with the LSTM model trained on the data set imputing with interpolation.