🏠 Melbourne Housing Price Prediction

📌 Project Overview

This project builds a machine learning model to predict housing prices using the Melbourne Housing dataset. It covers the full pipeline from data cleaning and feature engineering to model comparison and evaluation.

---

📊 Dataset

• Source: Melbourne Housing Snapshot
• Contains features such as:
  • Suburb, Rooms, Distance, Landsize, BuildingArea, YearBuilt, etc.
• Target variable: Price (log-transformed)

---

🔧 Data Preprocessing

• Handled missing values:
  • Filled Car and BuildingArea using median values
• Removed unrealistic entries:
  • Filtered YearBuilt > 1800
  • Removed extreme outliers in BuildingArea
• Feature Engineering:
  • HouseAge = CurrentYear - YearBuilt
  • Area_per_room = BuildingArea / Rooms
• Applied log transformation to target variable

---

🧠 Feature Engineering

• Used K-Fold Target Encoding for Suburb
  • Prevented data leakage
  • Handled unseen categories with fallback mean
• Removed original categorical columns
• Final feature set includes:
  • Rooms, Distance, Bathroom, Car, Landsize, BuildingArea
  • HouseAge, Latitude, Longitude, Propertycount
  • Suburb_encoded, Area_per_room

---

🤖 Models Used

🔹 Baseline Models

• Linear Regression
• Ridge Regression
• Lasso Regression

Performance:

• R² ≈ 0.75
• Provided a solid baseline but limited in capturing complex patterns

---

🌲 Random Forest

• Captures non-linear relationships and feature interactions

Performance:

• RMSE ≈ 0.21
• R² ≈ 0.85

---

⚡ XGBoost (Best Model)

• Boosting-based model with strong predictive performance

Performance:

• RMSE ≈ 0.175
• R² ≈ 0.895

---

📈 Model Comparison

Model	RMSE	R² Score
Linear	0.271	0.748
Random Forest	0.210	0.848
XGBoost	0.175	0.895

---

🔍 Key Insights

• Housing prices exhibit strong non-linear relationships
• Location (Suburb) is a critical feature
• Feature engineering significantly improves performance
• Tree-based models outperform linear models on structured data

---

🏁 Conclusion

• Final model explains ~89% of variance in housing prices
• XGBoost achieved the best performance
• Proper preprocessing and encoding were key to success

---

🚀 Future Improvements

• Hyperparameter tuning (GridSearch / RandomSearch)
• Try advanced boosting models (LightGBM, CatBoost)
• Deploy model using Streamlit or Flask
• Add more location-based or temporal features

---

🛠️ Tech Stack

• Python
• Pandas, NumPy
• Scikit-learn
• XGBoost
• Matplotlib, Seaborn

---

📌 Author

• Eshita Srivastava