Overview
This project forecasted online-channel product sales for the LG Aimers competition. The task was to predict future product demand from product sales history, product attributes, category hierarchy, and keyword-search-volume signals.
Goal
Predict multi-product sales over a fixed future horizon across multiple online distribution channels.
- Phase 1: July 2023
- Phase 2: August 2023
- Phase 3: September 2023 final evaluation
Approach
- Built a Python-based time-series forecasting workflow.
- Performed EDA on duplicate structures, product attributes, category hierarchy, sales history, and keyword-volume data.
- Evaluated LSTM-based models for product-level forecasting.
- Used a custom objective combining MSE with PSFA-style scoring considerations.
- Packaged the environment for reproducible model development.
Role
- Built the forecasting workflow and organized experiment notes for the LG Aimers Cartel team.
- Kept model trials, findings, and presentation material in a shared structure for final-stage collaboration.
Result
The work reached the LG Aimers online hackathon final stage and produced a reproducible forecasting pipeline for the competition setting.
Materials
The competition material summarized the product-sales forecasting task, final-stage workflow, and deliverables for LG Aimers.
