Comparative Study of PCA and ICA Based Traffic Flow Compression

Abstract

Data compression methods of traffic flow based on principle component analysis (PCA) and independence component analysis (ICA) were provided and compared. First, the statistical properties of freeway and urban road traffic flow were examined. It is found that the sample rate does not affect the analysis result and the distributions of deviation for traffic flow of different sample rates are similar. All of them are super-Gaussian distribution but near Gaussian distribution. Second, the compressions and reconfiguration based on PCA and ICA were tested and compared for freeway and urban road traffic flow respectively. The result shows that (1) PCA is a better choice for both traffic flows since they yield less non-Gaussian distribution; (2) the compression for freeway is better than for urban road since the freeway flow is more stable and regular. This result also gives some insight of the intrinsic nature of traffic flow.