House Price Prediction in Bangalore

Overview

• Goal: Predict price of a new house in Bangalore.
• Example:
  • Location: 6th Phase JP Nagar
  • Configuration: 4 BHK, 3 bathrooms, 2000 sqft
  • Predicted Price: 1 crore 43 lakhs INR

Tools and Technologies Used

• Flask: For building the web application
• Bootstrap: For designing the frontend
• JavaScript: To display predictions on the same page without redirecting

Dataset

• Source: Kaggle (Bengaluru house price data)
• File: bangalorehouse_data.csv
• Structure:
  • Columns:
    • Area Type
    • Availability
    • Location
    • Size (BHK)
    • Total Square Feet
    • Bathrooms
    • Balconies
    • Price (in lakhs)

Data Exploration

• Shape: 13,300 rows, 9 columns
• Data Types:
  • Area Type: categorical
  • Availability: categorical
  • Location: categorical
  • Size: categorical
  • Bathrooms: numerical
  • Price: numerical

Data Cleaning

1. Handling Missing Values:
   • Dropped columns: Area Type, Availability, Society, and Balcony due to high missing values.
   • Filled missing location values with 'Sarjapur'.
   • Filled missing size values with '2 BHK'.
   • Replaced bathroom nulls with median bathroom value.
2. Outlier Detection:
   • Cleaned total square feet values to fix ranges.
   • Dropped outliers based on price per square feet and BHK stats.
3. Feature Engineering:
   • Created price per square feet column.
   • Stripped leading/trailing whitespace from location strings and categorized locations with low counts as 'Other'.

Model Building

• Models Used:
  • Linear Regression
  • Lasso Regression
  • Ridge Regression
• Steps:
  • Imported necessary libraries
  • Trained models using cleaned dataset
  • Used pipelines for preprocessing and model fitting

Flask Application

1. Setup:
   • Created a basic Flask skeleton.
   • Implemented HTML template with Bootstrap for styling.
   • Form included fields for location, BHK, bathrooms, and total square feet.
2. JavaScript Integration:
   • Managed form submission without page reload using XMLHttpRequest.
   • Displayed prediction results dynamically.
3. Prediction Logic:
   • Loaded the trained model using Pickle.
   • Form input data was transformed into a DataFrame for predictions.
   • Displayed the prediction in lakhs, converted to rupees for final output.

Deployment

• Suggested using Flask-Cors for handling API requests in production (e.g., Heroku).

Conclusion

• Successfully built a web application to predict house prices in Bangalore based on user inputs.
• Achieved a good R2 score with the model, indicating predictive accuracy.