King County House Prices
Regression
Geospatial data
Predicting prices of 21,613 houses in the Seattle area, with geographic analysis (GeoPandas + Folium) and Ray Tune optimization.
Problem
Predict the sale price of 21,613 houses sold in King County (Seattle area) between 2014 and 2015.
Approach
Geographic analysis with GeoPandas and Folium, log transformation of the target and hyperparameter optimization with Ray Tune.
Technologies
Python GeoPandas Folium Ray Tune scikit-learn XGBoost Optuna SHAP
Results
XGBoost tuned with Ray Tune + Optuna reached R² 0.911 (log scale), MAPE 11.5% and MAE of US$64,191, versus 43.8% MAPE for the baseline and 19.9% for linear regression. SHAP analysis showed that waterfront, present in only 163 houses, raises the price by about 35% when it exists.