README.md

---
title: Zürich Apartment Price Predictor
emoji: 🏠
colorFrom: blue
colorTo: red
sdk: gradio
sdk_version: 6.8.0
app_file: app.py
pinned: false
short_description: Machine Learning Apartment Rent Price Predictor for Zurich
---

# Model Iterations Documentation
## Task: Apartment Price Prediction (Regression)

---

## Summary of Iterative Process

| Iteration | Objective | Key Changes | Models Used | CV Mean R² | CV Std Dev | Test MAE (CHF) | Fit Diagnosis |
|-----------|-----------|-------------|-------------|------------|------------|----------------|---------------|
| **1** | Build baseline model | - Basic cleaning<br>- 7 numerical features<br>- Train-test split (80/20)<br>- 5-fold CV | Linear Regression<br>Random Forest (n_estimators=100) | 0.429 (LR)<br>0.441 (RF) | 0.073 (LR)<br>0.075 (RF) | 549.90 (LR)<br>553.98 (RF) | ☐ Overfitting ☑ Underfitting ☐ Good Fit |
| **2** | Improve through feature engineering | - 7 new features created<br>- Feature scaling (Ridge)<br>- Hyperparameter tuning<br>- 5-fold CV | Ridge (alpha=10.0)<br>Random Forest (n_estimators=200, max_depth=15)<br>Gradient Boosting (n_estimators=150, max_depth=5) | 0.461 (Ridge)<br>0.481 (RF)<br>0.521 (GB) | 0.073 (Ridge)<br>0.070 (RF)<br>0.068 (GB) | 541.11 (Ridge)<br>544.92 (RF)<br>537.78 (GB) | ☐ Overfitting ☐ Underfitting ☑ Good Fit |

---

## Notes

**Metric:** MAE (Mean Absolute Error), R² (5-Fold Cross-Validation)

**Created Features:**
- rooms_per_sqm: Raumdichte (Zimmer pro m²)
- wealth_index: Wohlstandsindikator (tax_income × emp normalisiert)
- is_zurich_city: Binär-Feature für Zürich Stadt (PLZ 8000-8099)
- pop_emp_ratio: Bevölkerungs-Arbeitsplatz-Verhältnis
- log_area: Log-Transformation der Wohnfläche
- log_pop: Log-Transformation der Bevölkerung
- log_tax_income: Log-Transformation des Steuereinkommens

**Final Selected Features:**
- area (28.7% importance)
- log_area (25.0% importance)
- is_zurich_city (13.0% importance)
- rooms_per_sqm (12.2% importance)
- rooms, log_pop, pop_dens, pop, pop_emp_ratio, log_tax_income, emp, wealth_index, tax_income, frg_pct

**Final Model:** Gradient Boosting Regressor

**Reason for Selection:**
Best R² score (0.521) and lowest MAE (537.78 CHF). Consistent performance between CV and test set with lowest standard deviation (±67.62 CHF), indicating stable predictions.