|
Activity Number:
|
36
- Spatial and Spatio-Temporal Modeling of the Demographic and Economic Data
|
|
Type:
|
Invited
|
|
Date/Time:
|
Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
|
|
Sponsor:
|
Business and Economic Statistics Section
|
|
Abstract #316867
|
|
|
Title:
|
Tensor-Based Anomaly Detection from Large Spatio-Temporal Data
|
|
Author(s):
|
Taps Maiti and Peide Li*
|
|
Companies:
|
Michigan State University and Michigan State University
|
|
Keywords:
|
Anomaly Detection;
ADMM;
Tensor-decomposition;
Robust statistics;
Outlier detection
|
|
Abstract:
|
Detecting anomalies from traffic and other urban data has been one of the most popular topics in data science and machine learning studies. People are increasingly interested in knowing real-time traffic and potential hot-spots to make informed decisions on their day-to-day activities. Publicly available, high-quality, and fine-scale traffic volume measurements captured by various sensors carry a huge volume of information that can be utilized for detecting traffic anomalies such as accidents or traffic jams. The Taxi Data from New York City is an ideal data source with spatio-temporal complexity. It has great potential to provide information about hot-spots with heavy traffic in spatio-temporal locations. We develop a computationally efficient tensor-based spatio-temporal model for the traffic data to identify traffic hot-spots. We use an Alternating Direction Multiplier Method (ADMM) to estimate the model components.The numerical experiments indicate that the new framework can detect traffic anomalies with high accuracy.
|
Authors who are presenting talks have a * after their name.
