Breast Cancer Prediction: Classification Models and Neural Networks

This project compares two approaches for breast cancer prediction:

1. Approach 1: Traditional machine learning classification models (Logistic Regression, Random Forest, SVM, XGBoost).
2. Approach 2: Neural Networks using TensorFlow/Keras.

The goal is to evaluate the performance of both approaches on multiple breast cancer datasets and identify the best-performing model.

---

Datasets

The following datasets are used in this project:

1. Wisconsin Breast Cancer Dataset:

   • Features: Numerical (e.g., radius, texture, perimeter).
   • Target: Binary diagnosis (malignant/benign).

2. Breast Cancer Dataset:

   • Features: Numerical (e.g., age, tumor size) and categorical (e.g., menopause, metastasis).
   • Target: Binary diagnosis.

3. BRCA Dataset:

   • Features: Numerical (e.g., age, protein levels) and categorical (e.g., tumor stage, surgery type).
   • Target: Binary patient status.

4. German Breast Cancer Dataset:

   • Features: Numerical (e.g., age, tumor size) and categorical (e.g., hormonal status).
   • Target: Binary status.

5. SEER Breast Cancer Dataset:

   • Features: Numerical (e.g., age, survival months) and categorical (e.g., tumor stage, estrogen status).
   • Target: Binary status.

---

Approach 1: Classification Models

Models Used

• Logistic Regression
• Random Forest
• Support Vector Machine (SVM)
• XGBoost

Steps

1. Data Preprocessing:

   • Handle missing values.
   • Encode categorical variables.
   • Scale numerical features using StandardScaler.
   • Split data into training and testing sets.

2. Model Training:

   • Train each model on the training data.

3. Model Evaluation:

   • Evaluate models using Accuracy, ROC-AUC, and Confusion Matrix.
   • Combine predictions using a weighted average to create a fused model.

4. Visualization:

   • Plot ROC curves and bar plots for accuracy and AUC across datasets.

---

Approach 2: Neural Networks

Model Used

• A Sequential Neural Network with:
  • Input layer: 128 neurons, ReLU activation.
  • Hidden layers: 64 and 32 neurons, ReLU activation.
  • Output layer: Softmax (for multiclass) or Sigmoid (for binary classification).

Steps

1. Data Preprocessing:

   • Handle missing values.
   • Encode categorical variables.
   • Scale numerical features using StandardScaler.
   • Split data into training and testing sets.

2. Handle Class Imbalance:

   • Use SMOTE (Synthetic Minority Oversampling Technique) to balance the dataset.
   • Compute class weights for training.

3. Model Training:

   • Train the neural network using the balanced dataset.
   • Use Binary Crossentropy (for binary classification) or Categorical Crossentropy (for multiclass classification) as the loss function.
   • Optimizer: Adam with a learning rate of 0.001.

4. Model Evaluation:

   • Evaluate the model using Accuracy, ROC-AUC, and Confusion Matrix.
   • For multiclass classification, use Weighted ROC-AUC.

5. Visualization:

   • Plot accuracy and ROC curves for each dataset.

---

Evaluation Metrics

1. Accuracy

• Measures the proportion of correctly predicted instances.

2. ROC-AUC

• Measures the model's ability to distinguish between classes across all thresholds.

3. F1 Score

• Harmonic mean of Precision and Recall.

---

Results

• Approach 1 (Classification Models):

  • Best-performing model: XGBoost (highest accuracy and AUC).
  • Fused model (weighted average) outperforms individual models.

• Approach 2 (Neural Networks):

  • Achieves competitive performance, especially on imbalanced datasets.
  • SMOTE and class weighting improve model performance.

---

How to Run the Code

1. Clone the repository:

   git clone <repository-url>
   cd <repository-folder>

2. Install dependencies:

   pip install -r requirements.txt

3. Run the scripts:

   • For Approach 1 (Classification Models):

     python classification_models.py

   • For Approach 2 (Neural Networks):

     python neural_network.py

4. View results:

   • ROC curves and accuracy plots are saved in the results/ folder.

---

Dependencies

• Python 3.x
• Libraries:
  • pandas, numpy, scikit-learn, tensorflow, imblearn, matplotlib, seaborn, xgboost

---

Author

• Arpitha Thippeswamy

---