Breast Cancer Diagnosis Using R

Introduction

• We analyzed the Wisconsin Breast Cancer dataset (569 samples, 32 features) to predict benign vs malignant tumors.
• We applied feature selection to drop low-importance attributes before training.
• We implemented and compared several ML algorithms in R, focusing on accuracy, sensitivity, and specificity.

Dataset: https://www.kaggle.com/uciml/breast-cancer-wisconsin-data

Why R?

• Open-source language with strong statistical and visualization packages.
• Excellent ecosystem for data wrangling, plotting, and ML experimentation.

Modeling Pipeline

1. Pre-Processing

• 569 examples, 32 features.
• Used feature selection to rank features by importance and drop the least useful ones.

2. Classification Methods

We tested three classifiers:

Decision Tree

• Supervised ML algorithm where each node represents a feature and leaves represent outcomes.
• Commonly used for disease diagnosis (breast cancer, ovarian cancer, heart sounds, etc.).
• Metrics: Accuracy 92.1%, Sensitivity 93.1%, Specificity 90.68%.

Naive Bayes

• Based on Bayes’ theorem: P(A|B) = (P(B|A) * P(A)) / P(B).
• Metrics: Accuracy 93.8%, Sensitivity 95.69%, Specificity 92.67%.

KNN

• Classifies by majority vote of the k nearest neighbors.
• Metrics: Accuracy 96.67%, Sensitivity 99.47%, Specificity 92.29%.

3. k-Fold Cross-Validation

• Split the data into k folds (commonly k = 5 or 10).
• Train on k−1 folds and validate on the remaining fold, rotating through all folds.
• Larger k reduces bias by keeping training sets closer in size to the full dataset.

Findings

• KNN delivered the best balance of accuracy and sensitivity on this dataset.
• Feature selection improved model performance by removing noisy predictors.
• Cross-validation confirmed the robustness of the chosen hyperparameters.