Training a Naive Bayes Classifier for Text Categorization | CodeSignal Learn

Training a Naive Bayes Classifier for Text Categorization | CodeSignal Learn

Topic Overview and Objective

Hello again! Now that you are familiar with loading and preprocessing the dataset, today we'll learn how to train a Naive Bayes Classifier for text classification using Python. We'll be using the SMS Spam collection dataset in this example given below. By the end of this lesson, you will have a clear understanding about the principles of Naive Bayes algorithm and its implementation with the sklearn package in Python.

Train-Test Split Details

You have previously learned how to load our dataset and vectorize its messages, now it's time to split the data in two sets. The train set, where we already know the output that the model learns on, and the test set, where we test our model’s predictions on unseen data. We can leverage the sklearn package for that as done in the following code: Python from sklearn.model_selection import train_test_split # Split data into training and test sets X_train, X_test, y_train, y_test = train_test_split(X_tfidf, df['label'], test_size=0.25, random_state=42) The random state in our train-test split function ensures reproducibility. This is crucial so we invariably get the same split each time we run the code. Let's use a test size of 0.25, which means we will use 75% data for training and 25% for testing our model.

Introduction to Naive Bayes Model

The Naive Bayes classifier is a simple yet effective and commonly-used, probabilistic classifier. It's founded on applying Bayes' theorem with strong (naive) independence assumptions between the features. Naive Bayes classifiers have been particularly effective for high-dimensional data, and have worked quite well for text classification problems. In our case, we'll use the Multinomial Naive Bayes implementation available in sklearn, which is suitable for classification with discrete features (like word counts for text classification). Let's go ahead and create a Multinomial Naive Bayes model. Then, we can train it (or "fit" it, as we say in Machine Learning) with fit(X_train, y_train) . Here X_train contains the vectorized training data and y_train contains the corresponding labels. Python from sklearn.naive_bayes import MultinomialNB # Train a Multinomial Naive Bayes classifier clf = MultinomialNB() clf.fit(X_train, y_train)

Sign up

Join the 1M+ learners on CodeSignal

Be a part of our community of 1M+ users who develop and demonstrate their skills on CodeSignal

Predicting with a New Message

Once our model is trained, it can classify new, unseen messages. To do this, we preprocess and vectorize a new given message in the same way as our training data. Then, we use our trained classifier to predict the message category. Here's an example: new_message = "Congratulations! Text 'WIN' to claim a prize." new_message_vectorized = vectorizer.transform([new_message]) prediction = clf.predict(new_message_vectorized) print("The message is:", prediction) The message is: ['spam'] This shows how our trained Naive Bayes model can be used in practice to classify new messages.

Measuring Accuracy with Test Set

Upon training, our Naive Bayes classifier is now ready to classify new messages and measure its accuracy. To predict categories of unseen data, we employ the predict method on the vectorized test data, X_test . We then assess the model's performance via the accuracy metric - the proportion of correct predictions against the total predictions. The accuracy is calculated using accuracy_score() from sklearn's metrics. from sklearn.metrics import accuracy_score # Make predictions on the test set y_pred = clf.predict(X_test) # Calculate the accuracy of the predictions print("Accuracy:", accuracy_score(y_test, y_pred)) This outlines how our model fares in classifying SMS messages, providing insights into its reliability for practical applications. The output shows that the model has an accuracy of approximately 96%. This indicates that the Naive Bayes classifier is highly effective in distinguishing between spam and non-spam messages in our dataset, correctly identifying the category of new, unseen messages with a high degree of reliability. Accuracy: 0.9655419956927495

Python

new_message = "Congratulations! Text 'WIN' to claim a prize."
new_message_vectorized = vectorizer.transform([new_message])

prediction = clf.predict(new_message_vectorized)
print("The message is:", prediction)

Python

from sklearn.metrics import accuracy_score

# Make predictions on the test set
y_pred = clf.predict(X_test)

# Calculate the accuracy of the predictions
print("Accuracy:", accuracy_score(y_test, y_pred))