[MGT6705] Time Series Data Analysis and Forecasting Final-term Project
📝 Project Overview
This project aimed to forecast monthly Japanese tourist arrivals to Korea and identify the impact of Korea-related keyword trends on travel behavior. Using Google Trends data and actual tourist statistics, we developed interpretable models to enhance prediction accuracy and provide tourism policy insights. Served as a team leader, responsible for topic proposal, data analysis, and report writing.
🔬 Methodology
- Dataset: Monthly Japanese tourist arrivals (2010–2023) + 9 Korea-related search terms (e.g., “K-pop”, “Samgyeopsal”, “Myeongdong” in Japanese)
- Model: SARIMA & SARIMAX (with lagged external regressors)
- Time Series Decomposition: STL(Seasonal-Trend decomposition using Loess) to separate seasonal/trend components
- Model Selection: Grid search over 15,625 SARIMAX lag combinations
- Evaluation Metric: Root Mean Squared Error (RMSE)
- Final Selected Variables:
Samgyeopsal,K-pop,Myeongdong,Hangang,Dakgalbi(with optimized lags)
📈 Key Findings
- Samgyeopsal search interest precedes increases in tourists by ~3 months → strong long-term predictor.
- Myeongdong and Hangang show short-lag response (0–1 month), suitable for time-sensitive campaigns.
- SARIMAX models significantly outperformed SARIMA baselines in forecast accuracy.
- Seasonal lag patterns provide insight into how cultural and culinary interests shape Japanese tourist flow.
💡 Policy Recommendations
- 📍 Promote short-term campaigns for locations like Myeongdong with digital ads 1 month in advance.
- 🍲 Launch food-themed cultural promotions (e.g., Samgyeopsal Weeks) 2–3 months before peak seasons.
- 🎤 Leverage K-pop concerts (e.g., Waterbomb) to stimulate summer visits, traditionally a low-demand season.
These strategies allow tourism stakeholders to act on behavioral lead times reflected in online search trends.