Artificial Neural Network Modeling for Air Pollution Prediction: LSTM versus the Levenberg-Marquardt Approach

Goran Keković1   , Rade Božović1   , Sonja Ketin2   , Vladimir Mikić1   , Miloš Ilić1   , Boban Vesin3   

  1. Faculty of Information Technology, Alfa BK University, 11000 Belgrade, Serbia
    goran.kekovic@alfa.edu.rs (corresponding author), rade.bozovic@alfa.edu.rs, vladimir.mikic@alfa.edu.rs, milos.ilic@alfa.edu.rs
  2. School of Railroad Transport, Academy of Technical and Art Applied Studies, 11000 Belgrade Serbia
    sonja.ketin@vzs.edu.rs
  3. School of Business, University of South-Eastern Norway, Raveien 215, Borre, Vestfold, 3184, Norway
    boban.vesin@usn.no

Abstract

Accurate prediction of air pollutant concentrations remains a critical challenge for environmental monitoring and public health, demanding robust and adaptive artificial intelligence approaches. This study investigates the effectiveness of various types of artificial neural networks (ANNs), including Long Short-Term Memory networks (LSTM) and networks based on the Levenberg-Marquardt al gorithm (LM) and its variant with Bayesian regularization (LMBR), in predicting air pollution under different data conditions. Since LSTM networks are based on first derivative loss function algorithms and the LM algorithm is usually superior to this type of algorithm, a comparison of these networks was conducted. This is further supported by the limited coverage of this topic in the existing literature. ANNs were tested on two different datasets: the Air Quality dataset, where the target variable was the concentration of benzene (C6H6) and the Beijing PM2.5 dataset, where the target was the concentration of PM2.5 particles. The performance metrics of the ANNs were the root mean square error (RMSE), the mean absolute error (MAE), and the mean absolute percentage error (MAPE). It is shown that, in the case of the Air Quality dataset, the values of these parameters RMSE = 0.11 µg/m3, MAE = 0.09 µg/m3, MAPE = 1% for LSTM networks and RMSE = 0.14µg/m3, MAE = 0.092 µg/m3, MAPE = 1% for LMBR net works, were competitive. For LM networks, these values were significantly higher: RMSE = 0.57 µg/m3, MAE = 0.2 µg/m3, MAPE = 2%. Contrastingly, in the case of the Beijing database, the values of all parameters were drastically higher: RMSE ={45.74µg/m3, 64.94 µg/m3, 65.68 µg/m3}, MAE = {30.32 µg/m3, 42.83 µg/m3, 54.5 µg/m3} and MAPE = {52%,72%,74.25%} for LSTM, LMBR,andLMnetworks, respectively. In this case, the benzene concentration val ues exhibited a strictly linear correlation with the input variables. For the Beijing dataset, the relationships between PM2.5 concentration and predictor values were non-monotonic, leading to a drastic drop in the performance of all networks. Find ings reveal that, in this case, LSTM networks were more robust compared to LMBR and LM networks, as the values of their RMSE, MAE, and MAPE parameters were significantly lower. Furthermore, it is shown that the input variable selection tech nique (IVS) can be used to detect seasonal trends in the input data.

Key words

Artificial Intelligence, LSTM Network, Levenberg-Marquardt Algo rithm, Input Variable Selection.

Digital Object Identifier (DOI)

https://doi.org/10.2298/CSIS251101031K

Publication information

Volume 23, Issue 3 (June 2026)
Year of Publication: 2026
ISSN: 2406-1018 (Online)
Publisher: ComSIS Consortium

Full text

Download Available in PDF
Portable Document Format

How to cite

Keković, G., Božović, R., Ketin, S., Mikić, V., Ilić, M., Vesin, B.: Artificial Neural Network Modeling for Air Pollution Prediction: LSTM versus the Levenberg-Marquardt Approach. Computer Science and Information Systems, 23(3), 1077–1102 (2026). https://doi.org/10.2298/CSIS251101031K