/Breast_Canser_Wisconsin

Binary classification with Kaggle dataset

Primary LanguageJupyter Notebook

Breast Cancer Wisconsin (Diagnostic) Data Set

Kaggle dataset

GOAL: predict whether the cancer is benign or malignant.
BEST SCORE: 0.9956 AUC (best model - Logistic Regression)

Architecture

EDA

The features:

  1. ID number
  2. Diagnosis (M = malignant, B = benign)
  3. radius (mean of distances from center to points on the perimeter)
  4. texture (standard deviation of gray-scale values)
  5. perimeter
  6. area
  7. smoothness (local variation in radius lengths)
  8. compactness (perimeter^2 / area - 1.0)
  9. concavity (severity of concave portions of the contour)
  10. concave points (number of concave portions of the contour)
  11. symmetry
  12. fractal dimension ("coastline approximation" - 1)

image

Different models

K-Nearest Neighbors

image Total metrics:

Test (default) Test (7 neighbors)
Accuracy: 0.9649 0.9649
Precision: 0.9524 0.9524
Recall: 0.9524 0.9524
F1_score: 0.9524 0.9524

Logistic Regression

image Total metrics:

Test (default) Test (GridSearchCV)
Accuracy: 0.9766 0.9766
Precision: 0.9841 0.9841
Recall: 0.9538 0.9538
F1_score: 0.9688 0.9688

Decision Tree

Total metrics:

Test (default) Test (GridSearchCV)
Accuracy: 0.92 0.95

Feature importance: image

Best model

Best model is Logistic Regression tuned with GridSearchCV.