EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

Attention-Driven Hybrid CNN-LSTM with Dual Preprocessing for Missing Traffic Data Imputation in ITS

Yohanes Pracoyo Widi Prasetyo, - Linawati, Dewa Made Wiharta, Nyoman Putra Sastra

JOIV : International Journal on Informatics Visualization · 31.05.2026 · 10 (3) · 929

DOI: 10.62527/joiv.10.3.4422 ↗

Kurz gefasst

Fachartikel in der Zeitschrift JOIV : International Journal on Informatics Visualization, online erschienen am 31.05.2026. Verfasst von Yohanes Pracoyo Widi Prasetyo und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Missing traffic data is a critical challenge in Intelligent Transportation Systems (ITS), as sensor failures and communication network disruptions can severely degrade real-time traffic monitoring, analysis, and decision-making. This study proposes an attention-driven hybrid Convolutional Neural Network-Long Short-Term Memory (CNN–LSTM) model with a dual-stage preprocessing strategy to improve spatio-temporal data imputation under uncertain and noisy conditions. The preprocessing pipeline integrates a Kalman Filter to model and preserve temporal continuity, alongside a K-Nearest Neighbor (KNN) imputation method that refines spatial relationships across sensors, thereby reducing noise and stabilizing the input features before learning. The hybrid CNN–LSTM framework, augmented with a multi-head attention mechanism, enables the model to adaptively reweight spatio-temporal representations and focus on the most informative patterns in the data. Synthetic traffic datasets were generated via Monte Carlo simulations, with 20% of the data missing at randomly selected time intervals and locations. Model performance was evaluated using Kolmogorov–Smirnov, Mann–Whitney U, and Pearson correlation statistical tests. The results show that the proposed framework achieves low error rates (RMSE ≤ 2.10, MAE ≤ 1.70, MAPE < 3.6%) and strong correlations (R² > 0.94) across speed, volume, density, and temperature variables. A comparative analysis further demonstrates that the method outperforms conventional interpolation and machine-learning baseline approaches, particularly under high-missing-data conditions. These findings confirm the framework's robustness, scalability, and practical potential for real-time ITS operations, supporting more reliable traffic forecasting, congestion management, and intelligent mobility services.

Zitieren

APA

Prasetyo, Y. P. W., Linawati, Wiharta, D. M., & Sastra, N. P. (2026). Attention-Driven Hybrid CNN-LSTM with Dual Preprocessing for Missing Traffic Data Imputation in ITS. JOIV : International Journal on Informatics Visualization, 10(3), 929. https://doi.org/10.62527/joiv.10.3.4422

Vancouver

Prasetyo YPW, Linawati, Wiharta DM, Sastra NP. Attention-Driven Hybrid CNN-LSTM with Dual Preprocessing for Missing Traffic Data Imputation in ITS. JOIV : International Journal on Informatics Visualization. 2026;10(3):929. doi:10.62527/joiv.10.3.4422

BibTeX
@article{Prasetyo2026,
  author = {Prasetyo, Yohanes Pracoyo Widi and Linawati, - and Wiharta, Dewa Made and Sastra, Nyoman Putra},
  title = {Attention-Driven Hybrid CNN-LSTM with Dual Preprocessing for Missing Traffic Data Imputation in ITS},
  journal = {JOIV : International Journal on Informatics Visualization},
  year = {2026},
  volume = {10},
  number = {3},
  pages = {929},
  doi = {10.62527/joiv.10.3.4422}
}
Kostenlos starten →