This project demonstrates how a Neural Network can learn the XOR (Exclusive OR) problem, a classic example that cannot be solved using a single linear model. The implementation uses TensorFlow / Keras to train a simple feedforward neural network and visualize both the dataset and the learned decision boundary.
The XOR problem is a fundamental concept in machine learning that highlights the importance of hidden layers in neural networks. In this notebook:
A custom XOR dataset is loaded
A neural network with a hidden layer is trained
Model performance is evaluated
Decision boundaries are visualized
🧠 Model Architecture
Input Layer: 2 features (X1, X2)
Hidden Layer:
2 neurons
ReLU activation
Output Layer:
1 neuron
Sigmoid activation
Loss Function: Binary Crossentropy
Optimizer: Adam
The dataset (Xor_Dataset.csv) contains three columns:
Column Description X1 First input feature Y Second input feature Z XOR output label (0 or 1)
X Y Z 0 0 0 0 1 1 1 0 1 1 1 0
The notebook includes:
Scatter plot of the XOR dataset
Training loss curve
Decision boundary visualization showing how the neural network separates classes
Python
Pandas
NumPy
Matplotlib
TensorFlow / Keras
Why XOR cannot be solved with linear models
Importance of hidden layers in neural networks
How neural networks learn non-linear decision boundaries
Visual interpretation of model predictions
Sahil Bagri Aspiring Data Scientist / Machine Learning Enthusiast