New York city taxi analysis with graph signal processing

We apply graph signal processing to the study of taxi movement in New York City based on 2010–2013 New York City taxi data. Such analysis requires a signal extraction method that involves computing shortest paths between the start and end locations for each of the 700 million trip records. We perform spectral analysis on these graph signals, for which it is necessary to address the challenge of finding the eigendecomposition of the 6K-node directed Manhattan road network. We show that PSNR=29.90 dB is recovered for graph signals reconstructed from 70% of the graph frequency components. We illustrate that graph frequency components reveal taxi behaviors that are not obvious from the raw signal.