House Price Prediction using Neural Networks

This project demonstrates the implementation of neural network models to predict house prices using TensorFlow/Keras. The project explores different neural network architectures and techniques to improve model performance.

Project Overview

The project uses a dataset containing house price information with 10 input features. Three different neural network models are implemented to compare their performance:

1. Simple Neural Network

   • 2 hidden layers with 32 neurons each
   • ReLU activation
   • SGD optimizer

2. Deep Neural Network

   • 4 hidden layers with 1000 neurons each
   • ReLU activation
   • Adam optimizer

3. Deep Neural Network with Dropout

   • 4 hidden layers with 1000 neurons each
   • Dropout layers (0.5) after each hidden layer
   • ReLU activation
   • Adam optimizer

Dependencies

• pandas
• scikit-learn
• tensorflow/keras
• matplotlib

Dataset

The project uses housepricedata.csv which contains house-related features. The data is preprocessed using MinMaxScaler for normalization.

Model Architecture

All models are trained with:

• Binary cross-entropy loss
• Accuracy metric
• 100 epochs
• Batch size of 32
• Train/Validation/Test split (70%/15%/15%)

Visualizations

The project includes visualization of:

• Model loss curves (training and validation)
• Model accuracy curves (training and validation)

These visualizations help in understanding model performance and identifying potential overfitting/underfitting issues.

Usage

Run the Jupyter notebook housepricedata_NN.ipynb to:

1. Load and preprocess the data
2. Train the three different models
3. Visualize the results and compare model performance