Activity Number:
|
424
- Contributed Poster Presentations: Transportation Statistics Interest Group
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, July 30, 2019 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Transportation Statistics Interest Group
|
Abstract #304326
|
|
Title:
|
Weighted L1 Regularized VAR for Spatio-Temporal Data
|
Author(s):
|
Zhenzhong Wang* and Abolfazl Safikhani and Zhengyuan Zhu and David Matteson
|
Companies:
|
Iowa State University and Columbia University and Iowa State University and Cornell University
|
Keywords:
|
VAR;
spatio-temporal ;
LASSO;
forecast;
network detection
|
Abstract:
|
Vector Auto-regressive (VAR) model is commonly used to model multivariate time series. However in spatio-temporal data such as traffic network, the traditional high dimensional method does not include additional spatial and temporal information in model. This may lead to unreliable network detection and inaccurate forecast. We proposed a weighted l1 regularized approach which penalizes parameter differently according to the spatial distance and temporal lags in VAR. Its theoretical properties were explored in both exactly sparse case and weakly sparse case. The simulation study demonstrated its improvement over regular LASSO in both estimation and out-of-sample forecast. We also compared our method with LASSO in a traffic network dataset to show the advantages of proposed method in network detection and forecasting.
|
Authors who are presenting talks have a * after their name.