Making Sense of Human-generated Spatial-temporal Data from Urban Environment

Pan, Menghai

Etd

Making Sense of Human-generated Spatial-temporal Data from Urban Environment

Public

With the fast pace of global urbanization and the wide use of smart devices and GPS sets, there are massive human-generated spatial-temporal data in the urban environment, e.g., the trajectory data of the human-operated vehicles in the ride-hailing service or traditional taxi service. Human-generated spatial-temporal data can represent human behavior intrinsically, which enable us to understand human behavior in a data-driven fashion. Understanding human behavior can benefit people in many aspects. For example, understanding the decision-making process of taxi drivers can help them improve their operation efficiency, making sense of the behavior of the urban commuters can help urban planners better design the urban transportation system. In this dissertation, we propose and develop several novel machine learning techniques to help people make sense of the massive human-generated spatial-temporal data from urban environment. The ultimate goal of this dissertation is to help bridge the gap between real-world applications and laboratory researches. In particular, we try to deliver appropriate machine learning and statistical analysis solution frameworks for making sense of human-generated spatial-temporal data from urban environment in the following four aspects. 1. Understanding human learning curve can reward people in different ways, e.g., help the new learners improve their performance faster. The temporal dynamics of human behavior can reflect the learning curve of human beings, which makes it possible for us to understand human learning curve from their behavior data. In this topic, we propose data-driven approaches to understand what and how human agents learn over time. 2. Recent research demonstrates successes in learning human decision-making strategies from their behavior data using deep neural networks (DNNs). Such DNN-based models are "black box'' models in nature, making it hard to explain what knowledge the models have learned from human. To solve this problem, in this topic of the dissertation, we propose an explainable imitation learning framework to understand human behavior. 3. Identifying human agents from their behaviors is a significant task, which is helpful in many real-world applications, e.g., identifying drivers in ride-hailing service. In this topic, we propose the human mobility signature identification solution to identify human agents from their mobility data. 4. The significant achievements on developing autonomous vehicles stimulate us to envision the future transportation system with shared autonomous vehicles. In this topic, we employ the real-world demand and service data in current taxi system to study the feasibility of the future smart cloud commuting system.

Creator