COMPARISON OF LSTM AND GRU IN TRAFFIC DENSITY PREDICTION USING HISTORICAL VISUAL DATA AND YOLOV5 DETECTION
Journal of Engineering & Technological Advances · 30.06.2026 · 11 (1) · 44-67
DOI: 10.35934/segi.v11i1.163 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Engineering & Technological Advances, online erschienen am 30.06.2026. Verfasst von Filda Angellia und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
The problem of traffic congestion in urban areas requires adaptive and data-based predictive solutions. This study aims to develop vehicle density detection and prediction patterns based on historical visual data by integrating YOLOv5 object detection algorithms and sequential neural network models, namely Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). The study also compared the effectiveness of LSTM and GRU in capturing vehicle movement patterns over time. LSTM can be implemented on long-term prediction accuracy by preserving more complex historical contexts, while GRU can be implemented for better training efficiency and processing time due to its simpler architecture. The dataset consists of 254 videos and 13,435 frames with an average VPS of 10.0. The GRU model obtained an MSE value of 5.81, lower than LSTM 6.05, indicating that the mean squared difference between the prediction and the actual value was smaller in the GRU. The RMSE value in the GRU is 2.41, also lower than the LSTM 2.46. The difference of about 0.05 points, although it seems small, still shows consistency that the GRU prediction error is lower. In MAE, the GRU has a value of 1.84, lower than the LSTM 1.89, which means that the mean absolute difference between the prediction and the actual value is smaller in the GRU. This research not only contributes to intelligent transportation systems but also opens up opportunities to develop traffic monitoring systems that are responsive to vehicle dimensions and traffic environment dynamics in real-time.
Zitieren
APA
Angellia, F., Merlina, N., Subekti, A., & Handayanto, R. (2026). COMPARISON OF LSTM AND GRU IN TRAFFIC DENSITY PREDICTION USING HISTORICAL VISUAL DATA AND YOLOV5 DETECTION. Journal of Engineering & Technological Advances, 11(1), 44-67. https://doi.org/10.35934/segi.v11i1.163
Vancouver
Angellia F, Merlina N, Subekti A, Handayanto R. COMPARISON OF LSTM AND GRU IN TRAFFIC DENSITY PREDICTION USING HISTORICAL VISUAL DATA AND YOLOV5 DETECTION. Journal of Engineering & Technological Advances. 2026;11(1):44-67. doi:10.35934/segi.v11i1.163
BibTeX
@article{Angellia2026,
author = {Angellia, Filda and Merlina, Nita and Subekti, Agus and Handayanto, Rahmadya},
title = {COMPARISON OF LSTM AND GRU IN TRAFFIC DENSITY PREDICTION USING HISTORICAL VISUAL DATA AND YOLOV5 DETECTION},
journal = {Journal of Engineering & Technological Advances},
year = {2026},
volume = {11},
number = {1},
pages = {44--67},
doi = {10.35934/segi.v11i1.163}
}
EUVIMED