This project develops an LSTM-CNN-based deep learning model for predicting high-risk areas of fishing vessel maritime accidents.
Marine accident records were transformed into risk surfaces using Spatiotemporal Kernel Density Estimation, or ST-KDE, and used as the target variable. Marine traffic data, meteorological data, and marine geospatial information were used as explanatory features.
For model training, these variables were integrated into a grid- and month-level dataset.
The proposed model achieved improved performance across all evaluation metrics compared with XGBoost and standalone LSTM models, recording a maximum hotspot recall of 95.6%.
Scenario-based analysis was conducted to examine changes in accident risk under different conditions. In addition, a web dashboard was developed to allow users to input conditions and visualize corresponding changes in predicted risk levels.
The model is expected to support monthly safety management policies and operational decision-making for fishing vessels.
웹 대시보드의 경우, 배포시 서버 유지 관련 사유로 시연 영상으로 대체합니다.
문의: v2earts@naver.com
| Title | Date |
|---|---|
|
2026년 작품 업로드 기간 안내
|
2026.05.18 |
|
온라인 아카이브 이용 안내(저작권 및 개인정보 보호)
|
2026.05.18 |
|
2025년 작품 업로드 기간 안내
|
2025.05.14 |
|
2024년 작품 업로드 기간 안내
|
2024.05.20 |
|
2022년 작품 업로드 기간 안내
|
2022.05.18 |
© 2020-2026, 서울시립대학교 도시과학대학 온라인 아카이브.
All rights reserved. All content is protected by applicable laws.