Meeting with Professor Yoonsuk Lee

 

Key Outcomes

Jen and Professor Yoonsuk discussed the development of a real-time monitoring and forecasting system for vegetable prices (specifically eggplant, tomato, and a third crop) in Bukidnon, Mindanao 12. The system aims to provide an early warning mechanism for local government units (LGUs) to mitigate the impact of fluctuating fuel prices on food costs 34. Professor Yoonsuk will serve as a consultant, providing guidance on modeling and data standardization 5.

Project Scope and System Design

• Target Beneficiaries: Initially designed for LGU decision-making to implement social assistance programs for farmers 3.
• Pilot Area: A specific city or municipality in Bukidnon province will be used for the pilot test 6.
• Technical Components: The project includes purchasing a server for large-scale data storage and developing software/website platforms for data input and visualization 78.
• Key Variables: The primary focus is the effect of fuel prices on vegetable prices, with weather variables considered as potential secondary inputs 49.

Modeling and Data Strategy

• Data Standardization: Professor Yoonsuk emphasized the need to standardize measurement units (e.g., daily vs. weekly) to ensure monitoring data matches existing statistical datasets 1011.
• Model Simplification: Due to the target users (LGU officers), the team agreed to prioritize simple, functional models over overly sophisticated ones 12.
• Proposed Approaches:
  • Use of a baseline model (e.g., ARIMA) before moving to more complex time-series models 12.
  • Consideration of XGBoost as a strong forecasting alternative 13.
  • Potential benchmarking against the KREI (Korea Rural Economic Institute) forecasting models 14.
• Risk Metric: The system will incorporate a risk metric to categorize price alerts (e.g., green/red) for early warning 415.

Risks and Open Questions

• Data Granularity: There is a risk that aggregating high-frequency real-time data to match weekly/monthly statistics may cause specific data characteristics to disappear 16.
• Geographic Influence: The impact of prices from neighboring provinces on the local pilot area needs to be scoped 1718.
• Model Integration: The exact sequence and connection between different proposed models (e.g., LSTM, SVM) remain unclear and may need simplification 1319.

Action Items

• Jen: Gather available initial data to determine the input-output flow for the model 20.
• Jen: Clarify whether the system will be public or restricted to government users 2122.
• Professor Yoonsuk: Inquire with colleagues about how the KREI models separate forecasts for individual vegetables 23.
• Professor Yoonsuk: Provide lecture notes to Jen one week prior to the September 22nd session 24.