- Completed
- Data Analysis
- Time-Series Forecasting
- XGBoost
- LSTM
DIVE 2025
We built a data analysis and forecasting experiment system for transshipment volumes at the Port of Busan, winning first place in the Busan Port Authority track.
2025 · Completed
Forecasting transshipment volumes at the Port of Busan
At DIVE 2025, SkyDive took on the Busan Port Authority’s transshipment forecasting challenge. We began by exploring port cargo data and domestic and international economic indicators to identify useful predictors.
From preparation through on-site analysis, development, and presentation, the work earned first place in the Busan Port Authority track.
Understanding the domain before the data
The datasets combined different file formats and time ranges. We also needed to understand port terminology and measurement units. We checked data quality and relationships, and built a viewer to inspect samples from individual files.
Research papers, related analyses, and news helped us build context. Questions to the challenge sponsor and on-site mentoring clarified what the available data could tell us and what additional information we would need.
Testing different forecasting approaches
We compared statistical time-series models, machine learning, and deep learning, experimenting with SARIMA, XGBoost, and LSTM. We structured the code to make it easier to introduce new data and explanatory factors.
Expected correlations were difficult to establish at the monthly level, and some economic indicators had inconsistent effects over time. External data was not always available for the periods we needed. We treated these constraints as part of the analysis and adjusted our approach.
A system for incorporating new factors
Alongside model selection, we built a system for quickly testing newly discovered correlations and explanatory factors. During the event, we used mentoring to revisit the challenge’s objectives while working on analysis, implementation, and presentation materials.
Results and lessons
We won first place in the Busan Port Authority track, but did not place in the overall competition among track winners.
The project reinforced the importance of understanding the real question before choosing a model, identifying data limitations, and explaining results to people from other fields. Conversations with the sponsor helped us reconsider what was actually needed.
Behind the scenes
Seungmin Yang’s Korean-language Journal series covers the application, study and orientation, preparation, competition day, and awards. You can find it in the Korean Articles section.