Email Spam Filtering Using NLTK and Naive Bayes

Quick answer: Learn how to build a spam email classifier in Python using NLTK for text preprocessing and the Naive Bayes algorithm, with complete example code.

The Problem: Classifying Emails as Spam or Not Spam

Spam filtering is a text classification problem: given the content of an email, predict whether it is spam or legitimate, commonly called ham. It remains a classic, well understood introduction to NLP classification.

Step 1: Text Preprocessing

import nltk
import re
from nltk.corpus import stopwords
from nltk.stem import PorterStemmer

nltk.download('stopwords')
stop_words = set(stopwords.words('english'))
stemmer = PorterStemmer()

def preprocess(text):
    text = text.lower()
    text = re.sub(r'[^a-z\s]', '', text)          # remove punctuation and numbers
    words = text.split()
    words = [stemmer.stem(w) for w in words if w not in stop_words]
    return ' '.join(words)

sample = "Congratulations! You have WON a FREE prize, click now!!!"
print(preprocess(sample))

Lowercasing, removing punctuation, stripping common stopwords like "the" and "and", and stemming words to their root form all reduce noise so the model can focus on genuinely informative words.

Step 2: Converting Text to Numeric Features

from sklearn.feature_extraction.text import CountVectorizer

vectorizer = CountVectorizer()
X = vectorizer.fit_transform(cleaned_emails)   # a bag-of-words matrix

CountVectorizer converts text into a matrix of word counts, since Machine Learning algorithms need numeric input, not raw text.

Step 3: Training a Naive Bayes Classifier

from sklearn.model_selection import train_test_split
from sklearn.naive_bayes import MultinomialNB
from sklearn.metrics import accuracy_score, classification_report

X_train, X_test, y_train, y_test = train_test_split(X, labels, test_size=0.2, random_state=42)

model = MultinomialNB()
model.fit(X_train, y_train)

predictions = model.predict(X_test)
print(accuracy_score(y_test, predictions))
print(classification_report(y_test, predictions))

Why Naive Bayes Works Well for Spam Filtering

Naive Bayes assumes each word contributes independently to the probability of an email being spam, an assumption that is not literally true in language, but works surprisingly well in practice for text classification, and trains extremely fast even on large vocabularies.

Why Accuracy Alone Is Not Enough

Spam datasets are often imbalanced, with far more legitimate email than spam. A model predicting "not spam" for everything can still score high accuracy while being completely useless. Precision and recall specifically for the spam class matter more here, since missing genuine spam and blocking a legitimate email carry very different costs.

Classifying a New Email

new_email = ["Limited time offer, claim your reward now"]
cleaned = [preprocess(e) for e in new_email]
vectorized = vectorizer.transform(cleaned)

print(model.predict(vectorized))   # e.g. ['spam']

Common Interview Questions

Why does Naive Bayes work well for text classification despite its "naive" independence assumption?

Even though words are not truly independent in real language, the assumption simplifies computation dramatically, and in practice the relative probability rankings it produces are often accurate enough for reliable classification.

Why is accuracy a poor metric for a spam filter?

Spam datasets are typically imbalanced, so a model can achieve high accuracy simply by predicting the majority class for everything, while completely failing at the actual task of catching spam.