# PCNN: Deep Convolutional Networks for Short-term Traffic Congestion Prediction

**Authors:** Meng Chen, Xiaohui Yu, Yang Liu  
**arXiv:** [2003.07033](https://arxiv.org/abs/2003.07033)  
**Date:** 2020-03-16  
**Summary page:** https://www.summarizepaper.com/en/paper/2003.07033v1/

## Key points

- Traffic congestion affects quality of life and urban development
- Short-term traffic congestion prediction is crucial for planning daily activities and making informed decisions
- Meng Chen, Xiaohui Yu, and Yang Liu propose a novel method called PCNN based on deep CNNs to address these challenges
- PCNN uses time series folding and multi-grained learning to consider real-time traffic conditions and past traffic patterns
- PCNN models local temporal dependency and multiscale traffic patterns through convolutions over the input matrix
- The global trend of congestion can be addressed at the macroscale while more details and variations of congestion can be captured at the microscale
- PCNN outperformed baseline methods significantly in terms of accuracy when tested on a real-world urban traffic dataset
- PCNN provides an innovative solution for short-term traffic congestion prediction by utilizing deep CNNs to capture complex multiscale patterns in real-time traffic data effectively
- This research has important implications for improving transportation planning and management systems in urban areas worldwide.

## Summary

Traffic congestion is a major issue that affects the quality of life and urban development. Accurate short-term traffic congestion prediction is crucial for individuals and governments to plan their daily activities and make informed decisions. To address these challenges, Meng Chen, Xiaohui Yu, and Yang Liu propose a novel method called PCNN (Periodic Convolutional Neural Network) based on deep Convolutional Neural Networks (CNNs). PCNN uses two pivotal procedures: time series folding and multi-grained learning. It first folds the time series data into a two-dimensional matrix as the network input to consider both real-time traffic conditions and past traffic patterns. Then, with a series of convolutions over the input matrix, it models local temporal dependency and multiscale traffic patterns. The global trend of congestion can be addressed at the macroscale while more details and variations of congestion can be captured at the microscale. The authors tested their approach on a real-world urban traffic dataset and found that folding time series data into a two-dimensional matrix is effective for short-term congestion prediction. PCNN outperformed baseline methods significantly in terms of accuracy. In conclusion, PCNN provides an innovative solution for short-term traffic congestion prediction by utilizing deep CNNs to capture complex multiscale patterns in real-time traffic data effectively. This research has important implications for improving transportation planning and management systems in urban areas worldwide.