Applying Locally Linear Embedding on Feature Extraction of Traffic Flow Data

摘要

We encounter high dimensional data when using time series vector to describe the traffic flow of a certain link during some time period, or using flow data from different links to describe the traffic status of a region at a certain time point. This paper applies a dimensionality reduction method, named Locally Linear Embedding (LLE) to extract temporal and spatial features out of these high dimensional traffic flow data. LLE can visualize our data in a low dimension space, thus giving a vivid perspective on the emerging features. According to these features, we can put links into different clusters and better interpret the evolution of traffic patterns. Furthermore, comparison between linear dimensionality reduction method, PCA and LLE is carried out. The result shows that LLE has better performance.