Inferring High-Resolution Individual’s Activity and Trip Purposes with the Fusion of Social Media, Land Use and Connected Vehicle Trajectories
-
2017-11-30
-
By He, Qing
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report
-
Corporate Publisher:
-
Abstract:Trip purpose is crucial to travel behavior modeling and travel demand estimation for transportation planning and investment decisions. However, the spatial-temporal complexity of human activities makes the prediction of trip purpose a challenging problem. With the increasing advance of the Information Communication Technology (ICT), tremendous social media data becomes available. The goal of this report is to model and predict trip purpose with social media data. In order to achieve the goal of this report, first, this report provides a new approach to match Point of Interests (POIs) from Google Places API with Twitter data. Therefore, the popularity of each POI can be obtained. Moreover, a Bayesian Neural Network is employed to model the trip dependence within each individual’s daily trip chain and infer the trip purpose. In addition, to tackle the computational challenge in BNN, Elastic Net is implemented for feature selection before classification task. In addition, we also propose a Dynamic Bayesian Network for modeling and predicting trip purpose. Major findings are summarized as follows: We introduce a novel information retrieval method to match tweet with nearby Google Place Points of Interests (POIs) for trip prediction. The results show that our proposed method can reach up to 90% accuracy, whereas Foursquare tweet based method can only acquire 2%~16% accuracy. This study purposes a Dynamic Bayesian Network to model and predict trip purpose. Extensive experiments were conducted on real-world data sets, this method can achieve approximate 64% in average accuracy. This algorithm is more accurate when predict “shopping” activities, and the accuracy can be achieved as high as 80%. This report implements a feature selection method with Elastic Net. Total 29 features out of 45 are selected for modeling. The feature selection procedure is essential in a sense that it remarkably reduces the running time of BNN by 75%, from 60 minutes to 15 minutes.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: