Detecting Spatiotemporal Traffic Anomalies with Tensor Regression

A dynamic framework to model spatiotemporal traffic data to detect anomalous traffic data was proposed in this project. Instead of directly modeling the observed traffic data, we captured system dynamics and temporal dependencies by using a time-varying vector autoregressive (TVAR) model and applied a low-rank tensor structure to model the collection of evolving system matrices. The results show the superiority of the proposed model in uncovering anomalous traffic network dynamics and cannot be achieved by traditional temporal factorization models.

Anomaly Detection in Land Vehicle Traffic Activity

The objective of this project is to detect anomaly patterns and build a probabilistic model for a group of vehicles origin and destination based on historical GMTI track data and a priori road network. To address the problem, we applied probabilistic Tucker decomposition to uncover the latent patterns of spatial and temporal dimension and compared the expected values reconstructed from the latent patterns with observations to detect anomalies.

Forecasting Short-term Subway Ridership under Special Events Scenarios

The goal of this project is to predict short-term subway passenger flows under special events scenarios. We constructed the irregular fluctuation of Beijing subway ridership by using multiscale radial basis function (MSRBF) network based on limited passenger flow and applied fuzzy $c$-means algorithm and empirical functions to determine the parameters of MSRBF. To address the scalability problem, the matching pursuit orthogonal least squares algorithm was developed to produce a parsimonious model.

Epileptic Seizure Detection in EEGs

The aim of this project is to achieve an automatic detection framework of epileptic seizures from electroencephalography (EEG) signals. We developed an adaptive and localized time-frequency representation in EEG signals by multiscale radial basis functions. The extracted features of EEGs were fed into a support vector machine (SVM) to separate epileptic seizure from seizure-free EEG signals.