Projects   While most projects focus on South Korea, they can be applied and expanded to other cities and countries.

Transportation Informatics from Large Datasets

Identifying Functional Characteristics of Urban Areas Based on Travel Demand Patterns

Understanding the actual functions of places is a critical aspect of urban planning, transportation management, and regional development. While planned functions are determined by the built environment and land use, revealed functions result from the interaction between people and places, observable through travel demand patterns. Based on the assumption that places with similar activity types exhibit similar travel demand patterns, we propose a two-step data mining approach to identify revealed functions: i) identifying place clusters based on similar travel demand patterns for buses and taxis, and ii) inferring functional characteristics of each place cluster based on distinctive spatial context. Using travel demand data from buses and taxis in Daejeon Metropolitan City, South Korea, in 2019, we identify six place clusters, including residential, industrial, business/education/research, commercial, and mixed-functional areas. The goal is to provide insights into regional functions revealed by travel demand, emphasizing the significance of spatiotemporal features for informed urban decision-making.

This is a part of my Ph.D. dissertation.

Community Detection based on Origin-Destination Travel Demand

In network science, a community signifies a cluster of nodes with stronger interconnections than with nodes outside the cluster. Applying this idea to decipher travel demand patterns allows us to discern which regions or districts are strongly linked through people's travels, providing valuable insights for planning the living sphere for residents.

Association Rule-based Destination Analysis

We analyze destinations for Jeju tourists using the A Priori algorithm, a data mining technique. The A Priori algorithm identifies frequently appearing items and derives high-frequency item sets that include them. The results visualize the frequency-based connections between destinations and these analysis techniques and findings can be applied in the selection of stopping locations for tour buses.

Vehicle Route Density Visualization

Utilizing vehicle trajectory data*, we can trace vehicles departing from Daejeon City, South Korea. This allows us to understand the destinations people choose and the routes they take. Particularly, by visualizing density for each time frame, we can measure reachable areas within specific time periods, providing insights into the accessibility of inter-city travels.

* In this project, trajectory data is collected via a Dedicated Short Range Communication (DSRC) system.

Regularity in Vehicle Trips

The regularity of trips is a fundamental assumption for forecasting travel demand. In this project, one month of trajectory data from on-board transponders in vehicles in Daegu metropolitan city, South Korea, was used to unveil individual trip regularity at the trip chain level. Dynamic time warping and inter-spike interval algorithms were applied to quantitatively measure spatial and temporal regularity separately. The outcomes showed that i) trip regularity can be properly estimated using the indices, ii) spatial and temporal regularities are correlated - drivers who made trips at regular times also used similar paths in space across days, and iii) commuters and non-commuters have different distributions of regularity scores – commuters made more regular trips. This finding is intriguing as trip regularity can indicate the predictability of human mobility. Moreover, using regularity indices to measure historically collected trip behavior provides an alternative method for estimating or forecasting travel demand at the individual trip level.

Parking Demand Prediction

Limited parking information often leads to cruising-for-parking issues. To tackle this, forecasting parking availability based on spatio-temporal demand is crucial. We clustered multiple parking lots with similar temporal patterns and analyzed them spatially by building use. Using the Temporal Fusion Transformer, a transformer-based deep learning model, we achieved great prediction performance. Data from 41 off-street parking lots in Seoul revealed three clusters: leisure, residence, and business-related. Overall, prediction performance is great, but there are some differences depending on clusters. This project aims to provide accurate parking forecasts for efficient management and informed trip decisions.

Traffic Signal Control

Creating TOD Plans based on the Similarity of Traffic Characteristics

We propose a method for generating a TOD plan that reflects time-varying traffic characteristics using K-Means clustering. These traffic characteristics encompass turning rates for each entry-exit direction at a single intersection.

While the initial version focuses on a single intersection, we are now considering the development of an advanced method for coordination.