miniproject done 1

project_report.md

@@ -0,0 +1,101 @@
+# Real Estate Price Prediction Project  
+
+![Project Banner](https://education-team-2020.s3-eu-west-1.amazonaws.com/data-analytics/project+banners/real-state-project.jpg)
+
+## Overview  
+This project builds a machine learning model to predict house prices in King County, Washington based on property features. As a data scientist at a real estate firm, I developed a solution that:  
+- Processes 21,597 property records  
+- Identifies key price drivers through feature engineering  
+- Delivers predictions with $129,696 average error  
+- Provides actionable insights for business strategy  
+
+**Key Skills Demonstrated**: Python, Pandas, Scikit-Learn, Feature Engineering, Hyperparameter Tuning, Model Evaluation  
+
+## Dataset  
+Historic home sales (May 2014 - May 2015) with 21 features including:  
+- `price`: Target variable (sale price)  
+- `sqft_living`: Living area square footage  
+- `zipcode`: Location identifier  
+- `yr_built`: Year built  
+- `waterfront`: Waterfront view (binary)  
+- Full feature list available in `data/` directory  
+
+## Project Steps  
+
+### 1. Data Cleaning & Transformation  
+- Fixed temporal inconsistencies (e.g., houses "renovated after sale")  
+- Handled price outliers ($100K-$8M range) with log transformation  
+- Removed redundant features (e.g., `sqft_above` vs `sqft_living`)  
+
+```python
+# Fix illogical renovations
+df['yr_renovated'] = np.where(df['yr_renovated'] > df['sale_year'], 0, df['yr_renovated'])
+```
+
+### 2. Feature Engineering  
+Created 5 new predictive features:  
+1. `house_age`: 2024 - `yr_built`  
+2. `was_renovated`: Binary renovation flag  
+3. `zipcode_avg_price`: Neighborhood price baseline (*train-set only*)  
+4. `years_since_renov`: Years since last renovation  
+5. `price_to_zip_ratio`: Premium/discount vs neighborhood  
+
+### 3. Modeling & Evaluation  
+Tested 4 regression models:  
+| Model              | Test R² | Test RMSE   |  
+|--------------------|---------|-------------|  
+| Linear Regression  | 0.73    | $210,000    |  
+| Ridge Regression   | 0.73    | $209,800    |  
+| **Random Forest**  | **0.89**| **$129,696**|  
+| Gradient Boosting  | 0.87    | $135,400    |  
+
+### 4. Hyperparameter Tuning  
+Optimized Random Forest with GridSearchCV (108 combinations):  
+```python
+best_params = {
+    'n_estimators': 500,
+    'max_features': 'sqrt',
+    'min_samples_split': 2,
+    'min_samples_leaf': 1,
+    'max_depth': None
+}
+```
+Achieved +0.002 R² gain at $21K RMSE tradeoff  
+
+## Key Findings  
+
+### Top Price Drivers  
+1. **Location (ZIP Code)**: 45.3% impact  
+2. **Living Area Size**: 34.5% impact  
+3. **Geographic Coordinates**: 8.9% impact  
+4. **House Age**: 1.9% impact  
+5. **Waterfront View**: 0.9% impact  
+
+**Surprising Insight**: Renovation timing (`years_since_renov`) showed <1% impact - buyers value location more than updates!  
+
+### Business Recommendations  
+- Prioritize properties in premium ZIP codes over renovation projects  
+- Use model to identify undervalued homes in emerging neighborhoods  
+- Allocate 80% of improvement budget to feature engineering vs model tuning  
+
+
+## File Structure  
+```
+├── data/                    # Original datasets
+├── notebooks/               # Jupyter notebooks
+│   └── Real_Estate_Price_Prediction.ipynb  
+├── visualizations/          # Generated plots
+├── README.md                # This document
+```
+
+## Dependencies  
+- Python 3.12+  
+- Pandas, NumPy  
+- Scikit-Learn, XGBoost  
+- Matplotlib, Seaborn  
+- Jupyter Notebook  
+
+---
+
+**Presentation Summary**: [Download Slides](link_to_presentation.pdf)  
+**Final Model**: Random Forest Regressor (Test RMSE: $129,696)