California-Housing-Project

California Housing Analysis Project

This project follows a structured data science pipeline to analyze California housing data. Below is the step-by-step process based on the project roadmap.

Project Workflow

1. EDA (Exploratory Data Analysis)

I loaded the raw dataset and analyzed the distributions.

• Key Finding: I identified a "cap" in the housing prices. Any house worth more than $500,000 was recorded as exactly "5.0". This is an outlier that needs cleaning.

2. Train Baseline Linear Regression Model

I trained a simple Linear Regression model on the raw data to establish a baseline.

• Result: The model had a low accuracy score (~0.58), proving that a simple line cannot predict complex housing prices accurately.

3. Show Feature Importance

I analyzed which features influenced the baseline model the most.

• Insight: MedInc (Income) and Location (Latitude/Longitude) were the strongest predictors.

4. Feature Engineering

To improve the model, I applied several engineering techniques:

• Cleaning: Removed the "capped" rows (prices = 5.0) to reduce noise.
• Clustering: Used K-Means to group coordinates into 10 distinct geographical "neighborhoods".
• Synthetic Features: Created ratios like Bedroom_Ratio and People_per_Room to measure density and quality.

5. Train Regression Models (1, 2, 3)

I trained three models on the new processed data to predict the exact price:

• Model 1: Linear Regression (Improved).
• Model 2: Decision Tree Regressor.
• Model 3: Random Forest Regressor.
• 🏆 Winner: The Random Forest model was the best, achieving an R2 score of ~0.77.

6. Convert Regression → Classification Problem

I changed the problem definition from predicting the exact price to predicting a category.

• Method: Calculated the Median Price.
• Labels: Houses above median = 1 (Expensive); Houses below median = 0 (Cheap).

7. Train Classification Models (1, 2, 3)

I trained three classification models to predict "Expensive" vs. "Cheap":

• Model 1: Logistic Regression (Baseline).
• Model 2: Decision Tree Classifier.
• Model 3: Random Forest Classifier.
• 🏆 Winner: The Random Forest model won again with an accuracy of ~0.88.

---

Files in this Repository

• housing_regression_model.pkl: The trained Random Forest model for price prediction.
• housing_classification_model.pkl: The trained Random Forest model for classification.
• Assignment_2_Emilie_Levenbach.ipynb: The full Python code with all steps.

Presentation Video

https://www.loom.com/share/34429661f929403193c7b6bf29f93444

---

Created by Emilie Levenbach