Machine learning Web Application to detect Breast Cancer

Multiple predictive Machine Learning models are trained on Breast Cancer data for potentially ascertaining between malignant and benign tumor.

Dataset

Features are computed from a digitized image of a fine needle aspirate (FNA) of a breast mass. They describe characteristics of the cell nuclei present in the image. Also can be found on UCI Machine Learning Repository.

Attribute Information

1. ID number
2. Diagnosis (M = malignant, B = benign) 3-32)

Ten real-valued features are computed for each cell nucleus:

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

The mean, standard error and "worst" or largest (mean of the three largest values) of these features were computed for each image, resulting in 30 features. For instance, field 3 is Mean Radius, field 13 is Radius SE, field 23 is Worst Radius.

All feature values are recoded with four significant digits. lissing attribute values: none Class distribution: 357 benign, 212 malignant

Features

{
  "features": [
    "radius_mean",
    "texture_mean",
    "perimeter_mean",
    "area_mean",
    "smoothness_mean",
    "compactness_mean",
    "concavity_mean",
    "radius_worst",
    "texture_worst",
    "perimeter_worst",
    "area_worst",
    "smoothness_worst",
    "compactness_worst",
    "concavity_worst",
    "concave points_worst",
    "symmetry_worst",
    "fractal_dimension_worst"
  ]
}

• Minimum and Maximum values of each feature

  Feature	Min Value	Max Value
  radius_mean	6.981000	28.110000
  ractal_dimension_worst	9.710000	39.280000
  ymmetry_worst	43.790000	188.500000
  oncave_points_worst	143.500000	2501.000000
  oncavity_worst	0.052630	0.163400
  ompactness_worst	0.019380	0.345400
  moothness_worst	0.000000	0.426800
  rea_worst	7.930000	36.040000
  erimeter_worst	12.020000	49.540000
  exture_worst	50.410000	251.200000
  adius_worst	185.200000	4254.000000
  oncavity_mean	0.071170	0.222600
  ompactness_mean	0.027290	1.058000
  moothness_mean	0.000000	1.252000
  rea_mean	0.000000	0.291000
  erimeter_mean	0.156500	0.663800
  texture_mean	0.055040	0.207500

Models

• Logistic Regressor
• Support Vector Classifier
• Random Forest Classifier
• Decision Tree Classifier

Training

Below are the learning curves of models which did not undergo hyperparameter optimization.

Model	Learning Curves
Logistic Regressor
Support Vector Classifier
Random Forest Classifier
Decision Tree Clasifier

Hyperparameter Optimization

Using sklearn.model_selection.RandomizedSearchCV and sklearn.model_selection.GridSearchCV the parameter space for each model has been searched with parallel computing(n_jobs=-1).

{
  "logistic_regression_params": {
    "random_state": 0,
    "solver": "liblinear",
    "tol": 0.0001
  },
  "svc_params": {
    "C": 10,
    "gamma": 0.0001,
    "kernel": "rbf"
  },
  "random_forest_classifier_params": {
    "max_depth": 80,
    "min_sample_leaf": 4,
    "min_sample_split": 5,
    "n_estimator": 600
  },
  "decision_tree_params": {
    "criterion": "entropy",
    "max_depth": 3,
    "min_samples_leaf": 10
  }
}

Model	Learning Curves
Logistic Regressor
Support Vector Classifier
Random Forest Classifier
Decision Tree Classifier

Deployment

Requirements

numpy==1.21.4
pandas==1.3.4
scikit_learn==1.0.1
streamlit==1.3.0