Arduino Programming Classes for Kids in Nantes

Chapter 126 – Types of Naive Bayes

1. Choosing a Variant by Feature Type

Naive Bayes variants differ by how they model feature distributions. Match the model to your data:

• GaussianNB — continuous numbers (Gaussian / normal)
• MultinomialNB — discrete counts / frequencies (text, NLP)
• BernoulliNB — binary 0/1 presence–absence

2. Gaussian Naive Bayes

For continuous features (decimals, measurements). Each feature is modeled with a class-wise mean and variance (normal curve). Typical for medical or sensor data (e.g. Breast Cancer lengths).

3. Multinomial Naive Bayes

For integer counts (1, 2, 3…). Strong on Bag-of-Words and TF-IDF text features for classification and sentiment.

4. Bernoulli Naive Bayes

For binary features only (keyword present = 1, absent = 0). Classic for spam filters based on keyword presence.

5. Selection Guide

Variant	Feature type	Distribution	Best for
GaussianNB	Continuous	Normal (μ, σ²)	Medical, sensors, lengths
MultinomialNB	Counts	Multinomial	Text, BoW, TF-IDF
BernoulliNB	Binary 0/1	Bernoulli	Spam keyword flags

6. Python Sketch

from sklearn.naive_bayes import GaussianNB, MultinomialNB, BernoulliNB
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

X, y = load_breast_cancer(return_X_y=True)
X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.25, random_state=42)
gnb = GaussianNB().fit(X_tr, y_tr)
print(accuracy_score(y_te, gnb.predict(X_te)))  # ~0.937

Demo results (illustrative): GaussianNB ~93.7% on continuous cancer data; MultinomialNB on small text counts ~50%; BernoulliNB on binarized features ~56%. Always match the variant to the feature type for best results.