Gökçek, Öznur BegümDokuz, YeşimBozdağ, Aslı2024-11-072024-11-0720211866-7511https://doi.org/10.1007/s12517-021-08628-5https://hdl.handle.net/11480/11404In this study, PM10, SO2, NO2, NO, and NOX concentration values obtained for 2012–2018 from 7 different locations in Ankara city in Turkey were trained with deep learning systems, and predictions for the future were made. To make future predictions, time-based long short-term memory (LSTM) deep learning model was used. With the help of this model, it was predicted which values the air quality parameters determined in the city of Ankara for 2018 would take, and they were compared with the actual values of the same year. Accordingly, PM10 (R2 = 0.95, RMSE = 7.94), SO2 (R2 = 0.99, RMSE = 0.35), NO (R2 = 0.98, RMSE = 5.03), NO2 (R2 = 0.98, RMSE = 2.32), and NOX (R2 = 0.98, RMSE = 6.86) at almost all locations exhibited quite high performance for LSTM. According to the performance criteria obtained, it can be said that LSTM is useful in predicting air quality parameters and works successfully. © 2021, Saudi Society for Geosciences.eninfo:eu-repo/semantics/closedAccessAir pollutantNONO₂NO_xPM₁₀SO₂Time series LSTMArticle142110.1007/s12517-021-08628-52-s2.0-85118347711N/A