Coding and Robotics Classes for Kids in Lille

Chapter 127 – Build a Naive Bayes Classifier

1. Six-Step Pipeline

1. Load data and form a DataFrame
2. Inspect class balance; split 75% train / 25% test
3. Fit GaussianNB (continuous features)
4. Predict on the test set
5. Score accuracy, confusion matrix, classification report
6. Interpret FN vs FP in a medical context; optional single-sample inference

2. Dataset

Breast Cancer Wisconsin (Scikit-Learn): 569 samples, 30 continuous features (radius, texture, etc.), target 0 = malignant, 1 = benign. Class counts ≈ 212 malignant, 357 benign.

3. Code Skeleton

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
from sklearn.naive_bayes import GaussianNB
from sklearn.metrics import accuracy_score, confusion_matrix, classification_report

cancer = load_breast_cancer()
X_train, X_test, y_train, y_test = train_test_split(
    cancer.data, cancer.target, test_size=0.25, random_state=42)

model = GaussianNB()
model.fit(X_train, y_train)
y_pred = model.predict(X_test)

print(accuracy_score(y_test, y_pred))          # ~0.937
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred, target_names=cancer.target_names))

4. Results (illustrative)

Overall accuracy ≈ 93.71% on 143 test samples. Typical report: malignant precision ~0.96, recall ~0.87; benign precision ~0.93, recall ~0.98.

5. Confusion Matrix Focus

Example layout of outcomes: TN, FP, FN, TP. In this run, errors were small (e.g. 2 FN + 7 FP). In medical AI, false negatives (missed disease) are usually costlier than false positives (extra checks), so engineers often prioritize reducing FN.

6. Single-Sample Inference

sample = X_test[:1]
pred = model.predict(sample)
print(cancer.target_names[pred])  # e.g. ['benign']