HDB Resale Price Prediction (ML + Deep Learning)

End-to-end machine learning and deep learning pipeline for forecasting Singapore HDB resale prices using structured tabular and time-aware features.

This project is designed to showcase:

• data acquisition and preprocessing at scale,
• exploratory data analysis and feature engineering,
• sequence modeling with LSTM and GRU,
• hyperparameter tuning and artifact management,
• practical model comparison for real-world price prediction.

Project Goals

• Build a robust dataset from multiple public data sources related to HDB resale transactions and contextual features.
• Compare deep learning architectures (LSTM vs GRU) for predictive performance.
• Track preprocessing artifacts (scalers, vocabulary mappings, tuned model files) for reproducibility.
• Present a clear, portfolio-ready workflow from raw data to trained model outputs.

Repository Structure

.
├─ data/
│  ├─ raw/                         # Original source datasets
│  └─ processed/                   # Engineered/cleaned datasets used for modeling
├─ models/
│  ├─ lstm/                        # Saved LSTM model + preprocessing artifacts
│  └─ gru/                         # Saved GRU model + preprocessing artifacts + tuner logs
├─ notebooks/
│  ├─ 01_data_preprocessing_and_eda.ipynb
│  ├─ 02_lstm_baseline.ipynb
│  ├─ 03_lstm_hyperparameter_tuning.ipynb
│  ├─ 04_gru_baseline.ipynb
│  └─ 05_gru_hyperparameter_tuning.ipynb
└─ README.md

Data Assets

Raw Data (data/raw)

• HDB resale transaction datasets (multiple time ranges)
• HDB property information
• School information
• Train station reference data

Processed Data (data/processed)

• df_final.csv: merged feature table used in downstream workflows
• df_final_cleaned.csv: cleaned/transformed version for training

Modeling Workflow

1) Data Pre-analysis, Preprocessing, and EDA

Notebook: notebooks/01_data_preprocessing_and_eda.ipynb

Typical tasks covered:

• schema checks and missing-value profiling,
• dataset joining and consistency verification,
• exploratory visual analysis of price behavior,
• preprocessing decisions for model-ready inputs.

2) Baseline Modeling

Notebooks:

• notebooks/02_lstm_baseline.ipynb
• notebooks/04_gru_baseline.ipynb

Baseline model notebooks establish reference performance before tuning and provide an initial architecture comparison.

3) Hyperparameter Optimization

Notebooks:

• notebooks/03_lstm_hyperparameter_tuning.ipynb
• notebooks/05_gru_hyperparameter_tuning.ipynb

Tuning notebooks refine architecture/training parameters and save best-performing artifacts.

Saved Artifacts

The models/ directory stores reusable assets for inference and reproducibility:

• trained model files (.keras),
• feature and target scalers (.pkl),
• categorical vocabularies (.json),
• best hyperparameter configurations (.json),
• tuner search history (for GRU experiments).

Tech Stack

• Python
• Pandas, NumPy
• Scikit-learn
• TensorFlow / Keras
• Matplotlib, Seaborn

How To Run

Recommended: execute notebooks from the project root so relative paths are easier to manage.

1. Create and activate a virtual environment.
2. Install dependencies:

   pip install pandas numpy scikit-learn tensorflow matplotlib seaborn jupyter

3. Launch Jupyter:

   jupyter notebook

4. Run notebooks in this order:
   • 01_data_preprocessing_and_eda.ipynb
   • 02_lstm_baseline.ipynb
   • 03_lstm_hyperparameter_tuning.ipynb
   • 04_gru_baseline.ipynb
   • 05_gru_hyperparameter_tuning.ipynb

Author

This project is intended as a practical demonstration of applied machine learning and deep learning capability in a real estate pricing context.