AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORK
Many industries and companies in various fields are interested in time series analysis to predict the future. However, in time series modeling, precision is lacking as time progress. In this paper, an architecture is proposed, allowing on the one hand keep the prediction accurate over time using the...
Main Authors: | , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Copernicus Publications
2022-12-01
|
Series: | The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
Online Access: | https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLVIII-4-W3-2022/53/2022/isprs-archives-XLVIII-4-W3-2022-53-2022.pdf |
_version_ | 1811315857405509632 |
---|---|
author | Q. El Maazoui A. Retbi S. Bennani |
author_facet | Q. El Maazoui A. Retbi S. Bennani |
author_sort | Q. El Maazoui |
collection | DOAJ |
description | Many industries and companies in various fields are interested in time series analysis to predict the future. However, in time series modeling, precision is lacking as time progress. In this paper, an architecture is proposed, allowing on the one hand keep the prediction accurate over time using the Walk Forward Optimization (WFO); On the other hand, automate the choice of parameters of the statistical models (ARIMA) introducing “AutoArima”; The RNN models, especially LSTM architectures (LSTM, Bi-LSTM, Stacked LSTM) using the function Optuna. Moreover, to avoid overfitting the LSTM models, an automatic function is implemented in the presented architecture. To demonstrate the validity of this research, a comparison of three models applied to a railway company to predict the flow of passengers is made. In particular, the naive model constitutes a reference base, the ARIMA model which had demonstrated its performances in several research, and finally, following the last progress in the neural networks the LSTM architecture is introduced in the paper. According to the results, the implemented architecture has great potential and more accurate predictions by using WFO. Through the comparisons of the models made, Each model has proven its performance according to the case of study. More concretely the mean absolute error obtained by LSTM for the railway stations is 0,13 compared to 0,15 obtained by ARIMA and 0,16 for the naive model, showing a small superiority for LSTM over ARIMA. On the other side, ARIMA excels on the Train lines. |
first_indexed | 2024-04-13T11:38:38Z |
format | Article |
id | doaj.art-8f454cb33d0f44c6a55d68937e914d37 |
institution | Directory Open Access Journal |
issn | 1682-1750 2194-9034 |
language | English |
last_indexed | 2024-04-13T11:38:38Z |
publishDate | 2022-12-01 |
publisher | Copernicus Publications |
record_format | Article |
series | The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
spelling | doaj.art-8f454cb33d0f44c6a55d68937e914d372022-12-22T02:48:22ZengCopernicus PublicationsThe International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences1682-17502194-90342022-12-01XLVIII-4-W3-2022536010.5194/isprs-archives-XLVIII-4-W3-2022-53-2022AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORKQ. El Maazoui0A. Retbi1S. Bennani2Rime Team-Networking, Modeling and e-Learning Team, Masi Laboratory, Engineering.3S Research Center, Mohammadia School of Engineers (EMI), Mohammed V University in Rabat, MoroccoRime Team-Networking, Modeling and e-Learning Team, Masi Laboratory, Engineering.3S Research Center, Mohammadia School of Engineers (EMI), Mohammed V University in Rabat, MoroccoRime Team-Networking, Modeling and e-Learning Team, Masi Laboratory, Engineering.3S Research Center, Mohammadia School of Engineers (EMI), Mohammed V University in Rabat, MoroccoMany industries and companies in various fields are interested in time series analysis to predict the future. However, in time series modeling, precision is lacking as time progress. In this paper, an architecture is proposed, allowing on the one hand keep the prediction accurate over time using the Walk Forward Optimization (WFO); On the other hand, automate the choice of parameters of the statistical models (ARIMA) introducing “AutoArima”; The RNN models, especially LSTM architectures (LSTM, Bi-LSTM, Stacked LSTM) using the function Optuna. Moreover, to avoid overfitting the LSTM models, an automatic function is implemented in the presented architecture. To demonstrate the validity of this research, a comparison of three models applied to a railway company to predict the flow of passengers is made. In particular, the naive model constitutes a reference base, the ARIMA model which had demonstrated its performances in several research, and finally, following the last progress in the neural networks the LSTM architecture is introduced in the paper. According to the results, the implemented architecture has great potential and more accurate predictions by using WFO. Through the comparisons of the models made, Each model has proven its performance according to the case of study. More concretely the mean absolute error obtained by LSTM for the railway stations is 0,13 compared to 0,15 obtained by ARIMA and 0,16 for the naive model, showing a small superiority for LSTM over ARIMA. On the other side, ARIMA excels on the Train lines.https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLVIII-4-W3-2022/53/2022/isprs-archives-XLVIII-4-W3-2022-53-2022.pdf |
spellingShingle | Q. El Maazoui A. Retbi S. Bennani AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORK The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
title | AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORK |
title_full | AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORK |
title_fullStr | AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORK |
title_full_unstemmed | AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORK |
title_short | AUTOMATISATION HYPERPARAMETERS TUNING PROCESS FOR TIMES SERIES FORECASTING: APPLICATION TO PASSENGER’S FLOW PREDICTION ON A RAILWAY NETWORK |
title_sort | automatisation hyperparameters tuning process for times series forecasting application to passenger s flow prediction on a railway network |
url | https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLVIII-4-W3-2022/53/2022/isprs-archives-XLVIII-4-W3-2022-53-2022.pdf |
work_keys_str_mv | AT qelmaazoui automatisationhyperparameterstuningprocessfortimesseriesforecastingapplicationtopassengersflowpredictiononarailwaynetwork AT aretbi automatisationhyperparameterstuningprocessfortimesseriesforecastingapplicationtopassengersflowpredictiononarailwaynetwork AT sbennani automatisationhyperparameterstuningprocessfortimesseriesforecastingapplicationtopassengersflowpredictiononarailwaynetwork |