Breast Cancer (Sklearn)




1. 1. Load Data


feature names
import seaborn as sns
from sklearn.datasets import load_breast_cancer
data = load_breast_cancer()
print(data.feature_names)
['mean radius' 'mean texture' 'mean perimeter' 'mean area'
 'mean smoothness' 'mean compactness' 'mean concavity'
 'mean concave points' 'mean symmetry' 'mean fractal dimension'
 'radius error' 'texture error' 'perimeter error' 'area error'
 'smoothness error' 'compactness error' 'concavity error'
 'concave points error' 'symmetry error' 'fractal dimension error'
 'worst radius' 'worst texture' 'worst perimeter' 'worst area'
 'worst smoothness' 'worst compactness' 'worst concavity'
 'worst concave points' 'worst symmetry' 'worst fractal dimension']

30 variables et 569 points
X = data.data
print(f'{X.shape=}')
(569, 30)

2 classes : '0' (212 lignes) et '1' (357 lignes)
y = data.target
print('y: ', np.unique(y, return_counts=True))
(array([0, 1]), array([212, 357]))
Catégories sur mean_radius et mean_texture
plt.scatter(X[:, 0], X[:, 1], c=y, alpha=0.5, cmap='bwr')




2. 2. Train Test Split

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)
Train set avec 455 points et Test set avec 114 points
print(f'{X_train.shape=}')
X_train.shape=(455, 30)
print(f'{X_test.shape=}')
X_test.shape=(114, 30)
print(f'{y_train.shape=}')
y_train.shape=(455,)
print(f'{y_test.shape=}')
y_test.shape=(114,)



3. 3. KNeighborsClassifier

from sklearn.neighbors import KNeighborsClassifier

model = KNeighborsClassifier(n_neighbors=1)
model.fit(X_train, y_train)

train_predictions = model.predict(X_train)
test_prediction = model.predict(X_test)

train_accuracy = accuracy_score(y_train, train_predictions)
test_accuracy = accuracy_score(y_test, test_prediction)

Train Accuracy avec n_neighbors=1
train_accuracy = accuracy_score(y_train, train_predictions)
1.0

Test Accuracy avec n_neighbors=1
test_accuracy = accuracy_score(y_test, test_prediction)
0.9122807017543859



4. 4. Evaluation par Validation Croisée

from sklearn.model_selection import cross_val_score

model = KNeighborsClassifier(n_neighbors=1)
cv_scores = cross_val_score(
   estimator=model,
   X=X_train,
   y=y_train,
   cv=5,
   scoring='accuracy'
)

Scores de validation croisée
5 cv_scores (folds, splits)
[0.94505495 0.89010989 0.9010989  0.94505495 0.92307692]

Moyenne des scores de validation croisée
cv_scores.mean()
0.9208791208791209



5. 5. Recherche du meilleur Hyper-paramètre N_Neighbors

Validation croisée sur hyper-paramètre n_neighbors
k_scores = {}
for k in range(1, 50):
   model = KNeighborsClassifier(n_neighbors=k)
   cv_scores = cross_val_score(
       estimator=model, X=X_train, y=y_train, cv=5, scoring='accuracy'
   )
   k_scores[k] = cv_scores.mean()

plt.scatter(k_scores.keys(), k_scores.values())




6. 7. Optimisation de plusieurs Hyper-paramètres

param_grid = {
   'n_neighbors': range(1, 20),
   'metric': ['minkowski', 'cosine', 'euclidean']
}
grid_search = GridSearchCV(estimator=model, param_grid=param_grid, cv=5, scoring='accuracy')
grid_search.fit(X_train, y_train)
print(grid_search.cv_results_)

Contenu du dictionnaire
grid_search.cv_results_.keys
dict_keys(['mean_fit_time', 'std_fit_time', 'mean_score_time', 'std_score_time', 'param_metric', 'param_n_neighbors', 'params', 'split0_test_score', 'split1_test_score', 'split2_test_score', 'split3_test_score', 'split4_test_score', 'mean_test_score', 'std_test_score', 'rank_test_score'])

Paramètres testés
params = grid_search.cv_results_['params']
[{'metric': 'minkowski', 'n_neighbors': 1}, {'metric': 'minkowski', 'n_neighbors': 2}, {'metric': 'minkowski', 'n_neighbors': 3}, {'metric': 'minkowski', 'n_neighbors': 4}, {'metric': 'minkowski', 'n_neighbors': 5}, {'metric': 'minkowski', 'n_neighbors': 6}, {'metric': 'minkowski', 'n_neighbors': 7}, {'metric': 'minkowski', 'n_neighbors': 8}, {'metric': 'minkowski', 'n_neighbors': 9}, {'metric': 'minkowski', 'n_neighbors': 10}, {'metric': 'minkowski', 'n_neighbors': 11}, {'metric': 'minkowski', 'n_neighbors': 12}, {'metric': 'minkowski', 'n_neighbors': 13}, {'metric': 'minkowski', 'n_neighbors': 14}, {'metric': 'minkowski', 'n_neighbors': 15}, {'metric': 'minkowski', 'n_neighbors': 16}, {'metric': 'minkowski', 'n_neighbors': 17}, {'metric': 'minkowski', 'n_neighbors': 18}, {'metric': 'minkowski', 'n_neighbors': 19}, {'metric': 'cosine', 'n_neighbors': 1}, {'metric': 'cosine', 'n_neighbors': 2}, {'metric': 'cosine', 'n_neighbors': 3}, {'metric': 'cosine', 'n_neighbors': 4}, {'metric': 'cosine', 'n_neighbors': 5}, {'metric': 'cosine', 'n_neighbors': 6}, {'metric': 'cosine', 'n_neighbors': 7}, {'metric': 'cosine', 'n_neighbors': 8}, {'metric': 'cosine', 'n_neighbors': 9}, {'metric': 'cosine', 'n_neighbors': 10}, {'metric': 'cosine', 'n_neighbors': 11}, {'metric': 'cosine', 'n_neighbors': 12}, {'metric': 'cosine', 'n_neighbors': 13}, {'metric': 'cosine', 'n_neighbors': 14}, {'metric': 'cosine', 'n_neighbors': 15}, {'metric': 'cosine', 'n_neighbors': 16}, {'metric': 'cosine', 'n_neighbors': 17}, {'metric': 'cosine', 'n_neighbors': 18}, {'metric': 'cosine', 'n_neighbors': 19}, {'metric': 'euclidean', 'n_neighbors': 1}, {'metric': 'euclidean', 'n_neighbors': 2}, {'metric': 'euclidean', 'n_neighbors': 3}, {'metric': 'euclidean', 'n_neighbors': 4}, {'metric': 'euclidean', 'n_neighbors': 5}, {'metric': 'euclidean', 'n_neighbors': 6}, {'metric': 'euclidean', 'n_neighbors': 7}, {'metric': 'euclidean', 'n_neighbors': 8}, {'metric': 'euclidean', 'n_neighbors': 9}, {'metric': 'euclidean', 'n_neighbors': 10}, {'metric': 'euclidean', 'n_neighbors': 11}, {'metric': 'euclidean', 'n_neighbors': 12}, {'metric': 'euclidean', 'n_neighbors': 13}, {'metric': 'euclidean', 'n_neighbors': 14}, {'metric': 'euclidean', 'n_neighbors': 15}, {'metric': 'euclidean', 'n_neighbors': 16}, {'metric': 'euclidean', 'n_neighbors': 17}, {'metric': 'euclidean', 'n_neighbors': 18}, {'metric': 'euclidean', 'n_neighbors': 19}]

Scores obtenus
scores = grid_search.cv_results_['mean_test_score']
[0.92087912 0.91648352 0.93186813 0.92307692 0.92087912 0.92527473
 0.92747253 0.93406593 0.93626374 0.93406593 0.93186813 0.92967033
 0.93626374 0.93406593 0.92307692 0.92747253 0.92307692 0.92527473
 0.92087912 0.9010989  0.9010989  0.92087912 0.91648352 0.91868132
 0.91648352 0.92087912 0.91868132 0.92087912 0.92527473 0.92087912
 0.91868132 0.92527473 0.92087912 0.92307692 0.91868132 0.91868132
 0.91428571 0.91428571 0.92087912 0.91648352 0.93186813 0.92307692
 0.92087912 0.92527473 0.92747253 0.93406593 0.93626374 0.93406593
 0.93186813 0.92967033 0.93626374 0.93406593 0.92307692 0.92747253
 0.92307692 0.92527473 0.92087912]
grid_search_results = pd.DataFrame(params)
grid_search_results['score'] = scores
       metric  n_neighbors     score
0   minkowski            1  0.920879
1   minkowski            2  0.916484
2   minkowski            3  0.931868
3   minkowski            4  0.923077
4   minkowski            5  0.920879
5   minkowski            6  0.925275
6   minkowski            7  0.927473
7   minkowski            8  0.934066
8   minkowski            9  0.936264
9   minkowski           10  0.934066
10  minkowski           11  0.931868
11  minkowski           12  0.929670
12  minkowski           13  0.936264
13  minkowski           14  0.934066
14  minkowski           15  0.923077
15  minkowski           16  0.927473
16  minkowski           17  0.923077
17  minkowski           18  0.925275
18  minkowski           19  0.920879
19     cosine            1  0.901099
20     cosine            2  0.901099
21     cosine            3  0.920879
22     cosine            4  0.916484
23     cosine            5  0.918681
24     cosine            6  0.916484
25     cosine            7  0.920879
26     cosine            8  0.918681
27     cosine            9  0.920879
28     cosine           10  0.925275
29     cosine           11  0.920879
30     cosine           12  0.918681
31     cosine           13  0.925275
32     cosine           14  0.920879
33     cosine           15  0.923077
34     cosine           16  0.918681
35     cosine           17  0.918681
36     cosine           18  0.914286
37     cosine           19  0.914286
38  euclidean            1  0.920879
39  euclidean            2  0.916484
40  euclidean            3  0.931868
41  euclidean            4  0.923077
42  euclidean            5  0.920879
43  euclidean            6  0.925275
44  euclidean            7  0.927473
45  euclidean            8  0.934066
46  euclidean            9  0.936264
47  euclidean           10  0.934066
48  euclidean           11  0.931868
49  euclidean           12  0.929670
50  euclidean           13  0.936264
51  euclidean           14  0.934066
52  euclidean           15  0.923077
53  euclidean           16  0.927473
54  euclidean           17  0.923077
55  euclidean           18  0.925275
56  euclidean           19  0.920879
GridSearchCV Heatmap for KNN
grid_search_heatmap = grid_search_results.pivot(index='n_neighbors', columns='metric', values='score')
plt.figure(figsize=(10,8))
sns.heatmap(grid_search_heatmap, annot=True, fmt='.4f', cmap='viridis')
plt.title('GridSearchCV Heatmap for KNN')
plt.xlabel('Metric')
plt.ylabel('Number of neighbors')


Meilleurs paramètres
grid_search.best_params_
{'metric': 'minkowski', 'n_neighbors': 9}

Meilleur score
grid_search.best_score_
0.9362637362637363

Accuracy sur le jeu de test
best_model = grid_search.best_estimator_
y_pred = best_model.predict(X_test)
test_accuracy = accuracy_score(y_test, y_pred)
0.9649122807017544