Short-Term Prediction of Bus Station Fleet Number Using a Combination of BiLSTM Models

Predicting the number of bus station fleets requires a holistic approach, using sophisticated data analysis techniques and appropriate predictive modeling. Short-term predictions of bus station fleet numbers are proposed based on the best MAPE evaluation values ​​from the comparison of the Bi-LSTM,...

Full description

Saved in:
Bibliographic Details
Main Authors: Joko Siswanto, Ainun Rahmwati, Untung Rahardja, Nanda Dwi Putra, Muhammad Iman Nur Hakim, Tito Pinandita, Ilham Bagus Prasetyo
Format: Article
Language:English
Published: Universitas Muhammadiyah Magelang 2025-04-01
Series:Automotive Experiences
Subjects:
Online Access:https://journal.unimma.ac.id/index.php/AutomotiveExperiences/article/view/13402
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Predicting the number of bus station fleets requires a holistic approach, using sophisticated data analysis techniques and appropriate predictive modeling. Short-term predictions of bus station fleet numbers are proposed based on the best MAPE evaluation values ​​from the comparison of the Bi-LSTM, BiLSTM-CNN, BiLSTM-Transformer, BiLSTM-Informer, and BiLSTM-Reformer models. The dataset used is in the form of a CSV consisting of 6 types of arrivals and departures of the Giwangan City Yogyakarta type A bus station fleet from 01/01/2021 to 09/30/2023. The best prediction model was found in BiLSTM-Transformers based on a MAPE value of 0.2211 with a relatively fast time (00:00:52) compared to BiLSTM, BiLSTM-CNN, BiLSTM-Informer, and BiLSTM-Reformer. The BiLSTM-Transformer model can short-term predict 6 types of fleet arrivals and departures at the bus station in the next 30 days. The peak of the bar and curve is at 0 which means the proposed prediction model is very accurate. There is 1 strong positive correlation, 2 weak positive correlations, 2 strong negative correlations, 8 weak negative ones, and 2 uncorrelated ones. Prediction results can be used to support short-term decision making in fleet planning and management based on the dynamics of community mobility.
ISSN:2615-6202
2615-6636