Housing Price ANN

Predicting real estate with neural nets.

Real estate price prediction is a classic regression problem. This project applies a multi-layer Artificial Neural Network to a housing dataset, demonstrating how deep learning captures complex non-linear relationships in tabular data.

Deep learning on structured data.

The goal was to show that ANNs can compete with traditional regression when properly designed—ReLU activations, dropout regularization, early stopping, and the Adam optimizer all contribute to reliable generalization.

Neural network pipeline.

Pipeline includes dataset loading from sklearn, feature/target scaling with StandardScaler, a 128-64-1 ANN architecture with dropout, trained for 200 epochs with early stopping, evaluated on MSE.

Preventing overfitting on tabular data.

Neural networks on small tabular datasets (20k rows) overfit easily. Dropout (10%) and Early Stopping (patience=10) were essential — without them the validation loss diverged from training by epoch 50. The final model shows near-identical train/val curves, proof that deep learning can work on structured data with the right regularization.