House Price Predictor

⚡ Why This Project?

It’s a classic. Clean, tabular data; sensible preprocessing; baseline-to-better models; and clear evaluation with reproducible results. Not a passion project — just a tidy way to demonstrate end-to-end ML hygiene.

🎯 Objectives

• Data cleaning and feature engineering on a well-known housing dataset.
• Baseline models (Linear/ElasticNet) → tree-based models (RandomForest/XGBoost/LightGBM).
• Cross-validation, simple hyperparameter tuning, and holdout test evaluation.
• Interpretability: feature importances/SHAP summary for top drivers.
• Lightweight dashboard to show predictions and metric snapshots.

🧪 MVP Scope

• Notebook + scripts for ingest, preprocess, train, evaluate.
• 2–3 baseline models with k-fold CV; pick a champion by RMSE.
• Simple FastAPI endpoint for predict(); CSV upload input.
• Mini Next.js page to display RMSE/MAE/R² and top features.

🛠️ Tools & Stack (Planned)

• Python: pandas, scikit-learn, (optionally) xgboost/lightgbm
• FastAPI for a lightweight prediction service
• SQLite/PostgreSQL (optional) for experiment logs
• Next.js mini dashboard for metrics

📡 Data Notes

Use a widely used public housing dataset to keep the scope predictable and comparable. Focus is on clean preprocessing steps (missing values, categorical encoding, skew fixes) and reproducible evaluation.

📈 Metrics

• RMSE (primary), MAE, and R²
• Train vs validation gap (over/underfit check)
• K-fold stability (variance across folds)

🗺️ Roadmap (High Level)

1. Dataset setup + EDA
2. Preprocessing + baselines
3. Tree models + simple tuning
4. Interpretability (importances/SHAP)
5. FastAPI + mini metrics page