🏠 House Price Prediction using Linear & Ridge Regression

2026 • Suraj Singh
Data Science

cover

date: 2026-03-03

🏠 House Price Prediction using Linear & Ridge Regression

📌 Project Overview

This project demonstrates a complete end-to-end Machine Learning workflow for predicting house prices using Linear Regression and Ridge Regression.

The goal was not just to train a model, but to build a clean and professional ML pipeline including:

  • Proper data cleaning
  • Handling extreme outliers
  • Avoiding data leakage
  • Feature preprocessing using Pipeline & ColumnTransformer
  • Model comparison (Linear vs Ridge)
  • Alpha tuning for Ridge Regression
  • Prediction on unseen data

📂 Dataset

The dataset contains approximately 4600 house records with features such as:

  • Bedrooms
  • Bathrooms
  • Square footage (living & lot)
  • Floors
  • Waterfront
  • View
  • Condition
  • Year built / renovated
  • City
  • Price (Target Variable)

🧹 Data Cleaning

The following preprocessing steps were performed:

  • Dropped unnecessary and high-cardinality columns:

    • date
    • country
    • statezip
    • street

  • Removed extreme outliers using 99th percentile filtering:

    upper_limit = df["price"].quantile(0.99)
df = df[df["price"] <= upper_limit]

This stabilized the model performance significantly.

🔀 Train-Test Split

Data leakage was avoided by splitting the dataset before preprocessing:

X_train, X_test, y_train, y_test = train_test_split(
    X, y,
    test_size=0.2,
    random_state=42
)

🏗️ Preprocessing Pipeline

Used:

  • SimpleImputer (median for numeric, most_frequent for categorical)
  • StandardScaler (important for Ridge regularization)
  • OneHotEncoder(handle_unknown="ignore")
  • ColumnTransformer
  • Pipeline

This ensures clean, reproducible, production-style ML workflow.

📈 Model Performance

🔹 Linear Regression

  • R² Score ≈ 0.65
  • RMSE ≈ 171,000

🔹 Ridge Regression (Best Alpha = 0.01)

  • R² Score ≈ 0.65
  • RMSE ≈ 171,000

After proper cleaning, Linear Regression performed almost as well as Ridge, showing that the dataset became well-conditioned after outlier handling.

🎯 Key Insights

  • Linear Regression is sensitive to extreme outliers.
  • Removing top 1% price values significantly improved stability.
  • Ridge regularization helps when multicollinearity or large coefficients exist.
  • Proper preprocessing is more important than blindly switching models.
  • Avoiding data leakage is critical for trustworthy evaluation.

🏠 Unseen House Prediction Example

new_house = pd.DataFrame({
    "bedrooms": [3],
    "bathrooms": [2],
    "sqft_living": [1800],
    "sqft_lot": [4000],
    "floors": [1],
    "waterfront": [0],
    "view": [0],
    "condition": [3],
    "sqft_above": [1500],
    "sqft_basement": [300],
    "yr_built": [2005],
    "yr_renovated": [0],
    "city": ["Seattle"]
})

Predicted Price ≈ \$455,000

🚀 Tech Stack

  • Python
  • Pandas
  • NumPy
  • Scikit-Learn
  • Matplotlib

🧠 What This Project Demonstrates

  • Strong understanding of regression fundamentals
  • Regularization concepts (L2 Penalty)
  • Model evaluation (R² & RMSE)
  • Clean ML pipeline architecture
  • Real-world prediction workflow

📌 Conclusion

This project emphasizes building models the right way --- clean data, proper validation, and structured experimentation.

Instead of chasing unrealistic accuracy, the focus was on building a stable and interpretable regression pipeline.