Repository navigation
Methodology
The House Prices — Advanced Regression Techniques dataset (Ames, Iowa housing data) provides train.csv (with the SalePrice target) and test.csv (unlabeled, for the Kaggle-style submission).
A single engineered feature, TotalSqFt, sums three raw columns:
- Above-ground living area (
GrLivArea) - Total basement area (
TotalBsmtSF) - 1st + 2nd floor square footage
A plain LinearRegression (scikit-learn) is fit on three features: TotalSqFt, BedroomAbvGr (bedroom count), and FullBath (bathroom count), against the SalePrice target.
The model is evaluated on a held-out test split using RMSE (Root Mean Squared Error):
Root Mean Squared Error (RMSE): 46378.46
Diagnostic visualizations in the notebook include the target (SalePrice) distribution, a feature correlation heatmap, a predicted-vs-actual scatter plot, and a residuals plot — useful for sanity-checking that a three-feature linear model captures the dominant price signal (square footage) without fully modeling location, quality, and condition effects that a richer feature set would need.
The trained model's predictions on data/test.csv are written to a Kaggle-style data/submission.csv (Id, SalePrice columns).