<abstract xmlns="http://www.w3.org/1999/xhtml">

<p>Air pollution has increased worries regarding health and ecosystems. Precise prediction of air quality parameters can assist in the effective action of air pollution control and prevention. In this work, a deep learning framework is proposed to predict parameters such as fine particulate matter and carbon monoxide. Long Short Term Memory (LSTM) neural network-based model that processes sequences in forward and backward direction to consider the influence of timesteps in both directions is employed. For further learning, unidirectional layers’ stacking is implemented. The performance of the model is optimized by fine-tuning hyperparameters, regularization techniques for overfitting resolution, and various merging options for the bidirectional input layer. The proposed model achieves good optimization and performs better than the simple LSTM and a Recurrent Neural Network (RNN) based model. Moreover, an attention-based mechanism is adopted to focus on more significant timesteps for prediction. The self-attention approach improves performance further and works well especially for longer sequences and extended time horizons. Experiments are conducted using real-world data collected, and results are evaluated using the mean square error loss function.</p>
</abstract>

Air pollution has increased worries regarding health and ecosystems. Precise prediction of air quality parameters can assist in the effective action of air pollution control and prevention. In this work, a deep learning framework is proposed to predict parameters such as fine particulate matter and carbon monoxide. Long Short Term Memory (LSTM) neural network-based model that processes sequences in forward and backward direction to consider the influence of timesteps in both directions is employed. For further learning, unidirectional layers’ stacking is implemented. The performance of the model is optimized by fine-tuning hyperparameters, regularization techniques for overfitting resolution, and various merging options for the bidirectional input layer. The proposed model achieves good optimization and performs better than the simple LSTM and a Recurrent Neural Network (RNN) based model. Moreover, an attention-based mechanism is adopted to focus on more significant timesteps for prediction. The self-attention approach improves performance further and works well especially for longer sequences and extended time horizons. Experiments are conducted using real-world data collected, and results are evaluated using the mean square error loss function.

Long Short Term Memory Neural Network-Based Model Construction and Fne-Tuning for Air Quality Parameters Prediction

Department of CSE, Faculty of Technology and Engineering

Cybernetics and Information Technologies

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Air pollution has increased worries regarding health and ecosystems. Precise prediction of air quality parameters can assist in the effective action of air...

Long Short Term Memory Neural Network-Based Model Construction and Fne-Tuning for Air Quality Parameters Prediction

Published Online: Apr 10, 2022

Page range: 171 - 189

Received: Jul 28, 2021

Accepted: Dec 17, 2021

DOI: https://doi.org/10.2478/cait-2022-0011

Keywords
Air quality forecasting, Air pollution forecasting, Deep learning, Long short term memory, attention

© 2022 Virendra Barot et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Long Short Term Memory Neural Network-Based Model Construction and Fne-Tuning for Air Quality Parameters Prediction

Published Online: Apr 10, 2022

Page range: 171 - 189

Received: Jul 28, 2021

Accepted: Dec 17, 2021

DOI: https://doi.org/10.2478/cait-2022-0011

KeywordsAir quality forecasting, Air pollution forecasting, Deep learning, Long short term memory, attention

© 2022 Virendra Barot et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Keywords
Air quality forecasting, Air pollution forecasting, Deep learning, Long short term memory, attention