Problem
- Listings are everywhere, but fake listings are common and fair prices are hard to judge
- The midterm review pointed out the limits of RNN/LSTM models and a narrow data scope
University & Competitions
Busan Apartment Price Prediction
A desktop program that predicts apartment prices in Busan and maps prices and volume by district
Period | 2022.03 ~ 2022.06
Plenty of listings, but no way to tell a fair price
My Role
- Collected housing price and transaction data from the public data portal and preprocessed it by district (label encoding, log transforms)
- Compared LightGBM, XGBoost, RandomForest and others by RMSE, tuning LightGBM with 100 Optuna trials
- Visualized price trends and feature importance with Matplotlib, and district prices with folium maps
- Built a PyQt desktop UI to look up and compare predictions
Result
- A program offering an AI price estimate, yearly average price and volume, and time-series trends
- Narrowed the prediction scope from district to neighborhood level
