Austin House Price Range Prediction 🏡

This project aims to predict the price range of houses in Austin, TX, using various property attributes. The problem is framed as a multi-class classification task, where we classify each property into one of five predefined price brackets.

The repository showcases a complete machine learning workflow, including exploratory data analysis, feature engineering, model training, evaluation, and the incorporation of Natural Language Processing (NLP) features from property descriptions.

---

📋 Project Pipeline

The project follows these key steps:

1. Exploratory Data Analysis (EDA):

   • Analyzed numerical and categorical features to understand their distributions and relationships with the target variable (priceRange).
   • Visualized correlations between features, revealing strong multicollinearity between MedianStudentsPerTeacher, numOfBedrooms, and numOfBathrooms.

2. Feature Engineering:

   • Created a new feature, ageBuilding, from the yearBuilt column to represent the age of a property.
   • Dropped redundant or highly correlated features to improve model performance and reduce noise.

3. Model Training & Evaluation:

   • Baseline Model: A Logistic Regression model was trained to establish a performance baseline.
   • Primary Model: A LightGBM Classifier was implemented, which significantly outperformed the baseline.
   • Models were evaluated using Weighted F1-Score and Accuracy due to the multi-class nature of the problem.

4. Feature Importance:

   • Analyzed feature importances from the trained LightGBM model to identify the key drivers of house price ranges. Location (latitude, longitude), property age (ageBuilding), and school quality (avgSchoolRating) were identified as the most influential factors.

5. NLP Feature Integration:

   • Developed a text processing pipeline using TF-IDF Vectorization on the description column to extract valuable information and further enhance the model's predictive power. The text is cleaned by removing stop words and applying lemmatization.

---

📊 Results

The LightGBM model demonstrated superior performance compared to the Logistic Regression baseline.

Model	F1-Score (Weighted)	Accuracy
Logistic Regression (Baseline)	0.47	0.47
LightGBM Classifier	0.62	0.62

Key Predictive Features

The feature importance analysis highlighted that a property's price range is most influenced by:

1. Geographical Location (latitude & longitude)
2. Property Age (ageBuilding)
3. Average School Rating (avgSchoolRating)
4. Lot Size (lotSizeSqFt)

---

🚀 Getting Started

Prerequisites

Ensure you have Python 3.8+ installed. The required libraries are listed in requirements.txt. pandas numpy scikit-learn lightgbm matplotlib seaborn nltk joblib

Installation

1. Clone the repository:

   git clone [https://github.com/glopezzz/ML_challenge_FS.git](https://github.com/glopezzz/ML_challenge_FS.git)
   cd ML_challenge_FS

2. Install the required packages:

   pip install -r requirements.txt

3. Download NLTK data (required for NLP feature processing):

   import nltk
   nltk.download('stopwords')
   nltk.download('wordnet')

Usage

1. Run the Analysis: Open and run the live_exercise.ipynb notebook in a Jupyter environment to see the full analysis, training, and evaluation process.

2. Make Predictions: The trained model pipeline is saved in the /outputs directory. You can use the predict_price_range function to predict on new data.

   import pandas as pd
   import joblib
   
   # Load the trained model and label encoder
   model_pipeline = joblib.load('./outputs/lgbm_price_range_pipline.joblib')
   label_encoder = joblib.load('./outputs/label_encoder.joblib')
   
   # Create a sample DataFrame with new data (ensure columns match the training data)
   new_data = pd.DataFrame({
       'city': ['austin'],
       'homeType': ['Single Family'],
       'latitude': [30.26],
       'longitude': [-97.74],
       'garageSpaces': [2],
       'hasSpa': [True],
       'numOfPatioAndPorchFeatures': [1],
       'lotSizeSqFt': [8000.0],
       'avgSchoolRating': [8.5],
       'numOfBathrooms': [3.0],
       'ageBuilding': [15]
       # 'description' would be needed for the final NLP model
   })
   
   # Predict
   prediction_encoded = model_pipeline.predict(new_data)
   predicted_price_range = label_encoder.inverse_transform(prediction_encoded)
   probabilities = model_pipeline.predict_proba(new_data)
   
   print(f"Predicted Price Range: {predicted_price_range[0]}")
   print(f"Probabilities: {probabilities}")

---

🏛️ Repository Structure

. ├── data/ │ └── train.csv # Training dataset ├── outputs/ │ ├── lgbm_price_range_pipline.joblib # Saved model pipeline │ └── label_encoder.joblib # Saved label encoder ├── .gitignore ├── live_exercise.ipynb # Main Jupyter Notebook with all analysis └── README.md

🔮 Future Work

The notebook outlines several next steps to further improve the model:

• Full NLP Integration: Complete the training and evaluation of the model using the TF-IDF features from the property descriptions.
• Hyperparameter Tuning: Use GridSearchCV or a similar technique to optimize the hyperparameters of the LightGBM classifier and the TF-IDF Vectorizer.
• Model Deployment: Package the final model into a simple API (e.g., using Flask or FastAPI) for real-time predictions.