Large-Scale Location Mining for Multi-dimensional Risk Factors Discovery: A study for Road Accident Prevention

2014 Fiscal Year Research-status Report

• Project/Area Number: 26540166
• Research Institution: The University of Tokyo
• Principal Investigator: ホラノン ティーラユット 東京大学, 空間情報科学研究センター, 客員研究員 (00598287)
• Co-Investigator(Kenkyū-buntansha): 関本 義秀 東京大学, 生産技術研究所, 准教授 (60356087)
• Project Period (FY): 2014-04-01 – 2016-03-31
• Keywords: Traffic accidents / land use / GLM / zero-inflated models / risk factors

Outline of Annual Research Achievements

Summary of research achievements
- We successfully setup a collaborative research with our counterparts.- We successfully setup our cloud server and develop a script to process the data use for this research.- We concluded the data structure, sampling scheme, and identify possible risk factors use for this research.- We have studied and explored several analysis models and get our first results to determine the factors that highly related to traffic accidents. Our findings include:
Results of GLMs (Poisson regression, NB regression), and zero-inflated models (ZIP, ZINB) of traffic accident in weekday/weekend and holiday in Tokyo using　Land use data such as Densely inhabited area map, urban area map, land use map, elementary school counts POI data including number of banks,convenience store and shops, department stores, entertainment businesses, hospitals, high schools and universities, sport facilities, railway stations, tourist attractions Road data such as traffic volume, traffic speed, number of intersections, length of roads Note: the parameters have been selected individually for each model to get the best result Comparison between nested and non-nested models found the significance of zero-inflated models, and NB regression to have superiority over Poisson regression.The same data analysis for 32 other prefectures namely Hokkaido, Aomori, Miyagi, Fukushima, Ibaraki, Gunma, Saitama, Chiba, Kanagawa, Niigata, Ishikawa, Fukui, Nagano, Shizuoka, Aichi, Mie, Shiga, Kyoto, Osaka, Hyogo, Nara, Wakayama, Tottori, Shimane, Okayama, Tokushima, Kagawa, Kochi.

Current Status of Research Progress

2: Research has progressed on the whole more than it was originally planned.

Reason

We have submitted our first research finding to CUPUM2015 (Computers in Urban Planning and Urban Management) conference. CUPUM is a well-known peer-review international conference focusing on Urban Computing and the paper was accepted. The conference will be held at MIT, Boston, USA during July 7-10, 2015.

Strategy for Future Research Activity

Our plan for FY 2015, we will continue develop and test our analysis model to confirm the best solution (model adjustment) and find the most appropriated risk factors for road accident. After attending the international conference in July, we are planing to test the proposed model in various scenarios. We will also expand our model to test in other prefectures outside greater Tokyo. We are expected to submit our final results to a well-known international journal and writing a report at the end.

Causes of Carryover

Last fiscal year, we were focusing on the data exploration and model development to find the appropriated parameters for traffic risk factors. We spent less personal expenditure since the model have been testing only the pilot area. We expected to extend the proposed model for the entire Japan in the next fiscal year.

Expenditure Plan for Carryover Budget

The budget will be used for travel expense to attend the international conference this year (in July) also it will be used for hiring more research students to complete the second half of this research.