Spatio-temporal prediction of crime occurrence spots

DOI
  • Yaji Kaede
    nformation Engineering, National Defense Academy in Japan
  • Kubo Masao
    nformation Engineering, National Defense Academy in Japan
  • Sato Hiroshi
    nformation Engineering, National Defense Academy in Japan

説明

This paper proposes a method for spatiotemporal prediction of crime occurrence locations based on previous data. In recent years, Japanese government has begun to release data on crime occurrences to improve the efficiency of policing. In addition, the development of maps that can manage patrol and assist residents' crime prevention has been planned. For statistical crime prediction, while several methods are invented abroad, it has just begun to develop a specific crime prediction model for a low-crime country, Japan. One of the known methods uses LSTM to predict crime occurrences only from a temporal perspective, but it cannot predict points of crime occurrences and is insufficient to generate a map. Therefore, we propose a method that combines this LSTM based method with CNN that can adopt geographic locations. As a result of computer experiments, this method seems to be able to make predictions with a tendency to capture actual characteristics.

収録刊行物

キーワード

詳細情報 詳細情報について

  • CRID
    1390010292576683520
  • DOI
    10.5954/icarob.2022.os23-3
  • ISSN
    21887829
  • 本文言語コード
    en
  • データソース種別
    • JaLC
    • Crossref
    • OpenAIRE
  • 抄録ライセンスフラグ
    使用不可

問題の指摘

ページトップへ