Kaggle: House Price Prediction

This repository contains a complete machine learning project for predicting house sale prices, based on the Kaggle "House Prices: Advanced Regression Techniques" competition.

The project covers the entire data science workflow:

• Data Cleaning & Preprocessing: Handling missing values and feature engineering.
• Exploratory Data Analysis (EDA): Visualizing distributions and correlations.
• Model Training & Tuning: Building and optimizing multiple regression models.
• Model Interpretation: Using SHAP to understand why the model makes its predictions.

---

Final Model Performance

The best-performing model was a Tuned LightGBM (LGBM) Regressor.

Model	R² Score (Train)	R² Score (Test)	Test RMSE
Tuned LightGBM	0.9777	0.8832	$29,933.29

Note: RMSE (Root Mean Squared Error) is in dollars. R² measures the proportion of the variance in the sale price that is predictable from the features.

---

Tech Stack

• Data Analysis & Manipulation: pandas, numpy, scipy
• Machine Learning & Modeling: scikit-learn (Ridge, Lasso, RandomForest), xgboost, lightgbm
• Visualization & Interpretation: matplotlib, seaborn, SHAP
• Model Saving: joblib

---

Project Workflow

1. Data Cleaning & Preprocessing

• Missing Values: Categorical NaN values were filled with 'None', while numeric NaN values were imputed using the median or 0.
• Target Variable: The SalePrice was log-transformed (np.log1p) to normalize its distribution, which significantly improves regression model performance.
• Feature Scaling: Numeric features were scaled using StandardScaler and categorical features were encoded using OneHotEncoder.

2. Exploratory Data Analysis (EDA)

• Target Distribution: Analyzed the skewness of SalePrice and confirmed the effectiveness of the log transform.
• Feature Correlation: Used scatterplots and heatmaps to identify key features (like GrLivArea and OverallQual) that are highly correlated with the sale price.

3. Model Building & Tuning

Several regression algorithms were trained and evaluated using a Pipeline to streamline preprocessing:

• Ridge Regression (Tuned with GridSearchCV)
• Lasso Regression
• Random Forest Regressor
• XGBoost Regressor
• LightGBM Regressor (Best Performer)

4. Model Interpretation

"Why did the model predict this price?"

To answer this, a SHAP (SHapley Additive exPlanations) summary plot was created. This plot visualizes the impact of each feature on the model's output, showing which features are most important and whether high or low values of that feature increase or decrease the predicted price.

---

How to Run

1. Clone this repository:

   git clone [https://github.com/RaymussenArthur/House-Price.git](https://github.com/RaymussenArthur/House-Price.git)
   cd House-Price

2. Install dependencies: (It's highly recommended to create a requirements.txt file from your environment)

   pip install -r requirements.txt

3. Launch Jupyter Notebook:

   jupyter notebook

4. Open the main notebook to see the full analysis and model training process.