In order to join virtual sessions, you must be registered and logged-in(Were you registered for the in-person meeting in Denver? if yes, just log in.) 
Note: All session times are in Mountain Daylight Time.

Prediction model for the spatially spreading process within Itaewon commercial district using reinforcement learning

Authors: JIWON JANG*, Kyunghee University, Jinmu Choi, Kyung Hee University
Topics: Urban and Regional Planning, Urban Geography, Geographic Information Science and Systems
Keywords: Reinforcement learning, spatially spreading process, online popularity, gentrification, commercialization
Session Type: Poster
Day: 4/8/2020
Start / End Time: 3:20 PM / 4:35 PM
Room: Virtual Track 3
Presentation File: No File Uploaded


The purpose of this study is to create a model that predicts the dynamics of commercial area within Itaewon Commercial District using reinforcement learning algorithms. Itaewon Commercial District, one of the most important commercial districts in Seoul, is a research area. Itaewon Commercial District was initially formed around Itaewon subway station, however the commercial district have later expanded to residential areas such as Haebangchon, Gyeongridan-gil and Hannam-dong near Itaewon. The commercialization of residential areas due to the spread of Itaewon commercial districts led to an increase in residential costs and gentrification has occurred. To cope with such problems, reinforcement learning is applied to find areas where the community district could expand. Reinforcement learning is a learning algorithm that learns the optimal behavior policy according to the behavior of agents interacting with the environment. In order to learn about the spatially spreading process in the commercial district, three administrative districts covering the Itaewon commercial district are generated as grid worlds with 50m resolution in which agent can interact. Agent can stay or move in eight directions, including north, south, west, east, northwest, northwest, southwest, and southwest on the environment. As the simulation episode progresses, the agent can derive spatially different Q values for individual grids. As a result of reinforcement learning simulation, we looked for policy implications to respond to gentrification. Also, it is meaningful to explore the possibility that reinforcement learning was applied not only to the environment such as games, but also to the real world problems.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login