Breast Cancer Classification with Neural Networks

This project implements a binary classification model to predict breast cancer diagnosis (malignant vs benign) using neural networks with Keras/TensorFlow.

Dataset

The project uses the cancer.csv dataset containing breast cancer diagnostic features:

• Features: 30 numerical features including mean radius, texture, perimeter, area, smoothness, compactness, concavity, etc.
• Target: Binary classification (1 = Malignant, 0 = Benign)
• Size: Multiple samples with comprehensive feature measurements

Model Architecture

• Input Layer: 30 features
• Hidden Layers: 2 dense layers with 100 neurons each (ReLU activation)
• Output Layer: 1 neuron with sigmoid activation for binary classification
• Optimizer: Adam
• Loss Function: Binary crossentropy

Implementation

Data Preprocessing

• Train/test split (75%/25%)
• Feature standardization using StandardScaler
• Random state = 2 for reproducibility

Training

• Epochs: 10
• Batch size: 16
• Validation data: Test set

Visualization

• Training/validation accuracy curves
• Training/validation loss curves

Files

• Week_2.ipynb: Main notebook with complete implementation
• data/cancer.csv: Breast cancer dataset
• README.md: Project documentation

Requirements

Create a requirements.txt file with:

pandas>=1.3.0
numpy>=1.21.0
matplotlib>=3.4.0
scikit-learn>=1.0.0
tensorflow>=2.8.0

Installation

1. Clone the repository
2. Install dependencies:

pip install -r requirements.txt

Usage

1. Create a data/ folder and place cancer.csv inside it
2. Mount Google Drive (if using Colab)
3. Load and preprocess the dataset from data/cancer.csv
4. Train the neural network model
5. Evaluate performance with accuracy/loss plots

Expected Results

The model achieves the following performance metrics:

• Training Accuracy: ~95-98%
• Validation Accuracy: ~92-96%
• Training Loss: Decreases steadily over epochs
• Validation Loss: Should remain stable without significant overfitting

The model tracks both training and validation metrics to monitor performance and detect potential overfitting patterns.