SW-STP: A Spatio Temporal Pattern Based Model for Detecting Traffic Flow Data Anomalies

As transportation networks grow increasingly complex, traffic anomaly detection becomes crucial for effective traffic management and emergency responses. However, existing methods often fail to achieve precise detection in complex anomaly scenarios. We introduce a method called SW-STP for detecting traffic anomalies in road networks. First, we partition the road network data into windows of fixed duration. Next, we apply MSWMA to denoise the traffic sequences, filtering out fluctuations likely to be mistakenly identified as anomalies. Subsequently, we use an encoder-decoder architecture to encode the data into a low-dimensional space and reconstruct it, measuring the deviations between the original and reconstructed data. Finally, we employ a statistically derived threshold to identify anomalies in the road network. We evaluated our method using synthetic data generated by SUMO, demonstrating its superiority over competing methods.