Autoencoder-based Continual Outlier Correlation Detection for Real-Time Traffic Flow Prediction

In urban landscapes, traffic congestion, often identified by outlier events like accidents or constructions, poses a significant challenge. These outliers result in abrupt traffic fluctuations, necessitating real-time modeling for accurate traffic predictions. The proposed Outlier Weighted Autoencoder Modeling (OWAM) framework addresses this by employing autoencoders for local outlier detection at each traffic sensor and generating correlation scores to assess neighboring traffic's impact. These scores, which serve as the weighted information of the neighboring sensors, enhance the model's performances and enable effective real-time updates. OWAM achieves a balance between accuracy and efficiency, making it highly suitable for real-world applications. This advancement in traffic prediction models significantly contributes to the field of transportation management. The framework and its datasets are publicly available under https://github.com/himanshudce/OWAM.