📧 Classification · Naïve Bayes

Spam or Not?

Label emails, build a word-frequency table, and train a real Naïve Bayes classifier — the same algorithm that powers Gmail's spam filter!

📧 Label Emails

📊 Word Frequencies

🧪 Test Your Filter

📈 Accuracy Report

🏆 Badge

How Naïve Bayes Works

📧

Collect Examples

Label emails as Spam or Ham (not-spam). These become your training data.

🔢

Count Word Frequencies

Count how often each word appears in spam vs ham emails. "Free" appears more in spam!

📊

Calculate Probabilities

P(spam|email) = P(word1|spam) × P(word2|spam) × … × P(spam). Bayes' theorem!

⚖️

Classify

Whichever probability is higher — spam or ham — wins. That's the prediction!

📧

Wizzy the AI Tutor

Welcome to the Spam Lab! 🎉 Read each email carefully and label it Spam 🔴 (junk, suspicious, too good to be true) or Ham 🟢 (real, useful email). Your labels teach the Naïve Bayes model what spam looks like!

Step 1 — Label the Training Emails

0 / 12 labelled

📊 Labelling Progress

Labelled0 / 12

Spam0

Ham0

Remaining12

💡 Spam clues: Urgent language, prize offers, "FREE", suspicious senders, requests for money or passwords.

📧

Wizzy the AI Tutor

Your labels taught the model something! Now look at which words appear most in spam vs ham. Words like "free", "winner", "urgent" are far more common in spam. The model uses these frequencies to calculate the probability that a new email is spam!

Step 2 — Word Frequency Analysis

🔴 Top Spam Words

🟢 Top Ham Words

📧

Wizzy the AI Tutor

Time to test your trained spam filter! 🧪 For each new email, the model calculates P(spam) and P(ham) and picks the higher one. Watch the probability bars and see which words pushed the decision. Can the filter catch all the spam?

Step 3 — Test the Spam Filter

0 / 8 tested

What do YOU think?

🤔

Classify an email to see probabilities

🔴 Spam

—

🟢 Ham

—

Key evidence words

Classify an email first

🧪 Test Results

Tested0 / 8

Correct0

Wrong0

Your accuracy—

📧

Wizzy the AI Tutor

Look at the full accuracy report! 📈 Precision = how many spam predictions were actually spam. Recall = how many real spams did we catch. F1 score combines both. Real email filters aim for 99%+ — but they use millions of emails for training!

Accuracy Report

—

Accuracy

—

Precision

—

Recall

—

F1 Score

📧

Wizzy the AI Tutor

🎊 You've just built a real Naïve Bayes spam classifier — the same algorithm behind Gmail, Outlook, and Yahoo Mail! You understand training data, word frequencies, probability, and evaluation metrics. That's real machine learning engineering!

📧

Spam Hunter Badge!

You trained a Naïve Bayes classifier from scratch and evaluated its performance!

Optional. Stays on this device only — not sent to WhizzStep.

📧 WhizzStep AI Lab

Activity completion card for

Student Name

has built and tested a Naïve Bayes Spam Classifier

This records completion of a browser activity only. It is not an accredited certificate or proof of mastery.

Spam Hunter

Naïve Bayes Expert

ML Engineer

Accuracy: — · whizzstep.in

Key Concepts Mastered

Naïve Bayes

🎲 The Algorithm

Assumes each word contributes independently to spam probability. "Naïve" because real words aren't independent — but it still works great!

Prior Probability

📊 Base Rate

P(spam) = fraction of spam in training data. If 6 of 12 emails are spam, P(spam) = 0.5.

Laplace Smoothing

➕ Handling Zeros

What if a word never appeared in training? We add +1 to every count so probabilities are never zero.

Precision vs Recall

⚖️ The Trade-off

High precision = few false alarms. High recall = catch all spam. F1 score balances both!

Log Probabilities

🔢 Avoiding Underflow

Multiplying many tiny probabilities gives near-zero. We add log probabilities instead — same result, no underflow!

Real World

🌍 Where It's Used

Gmail, Outlook, Yahoo Mail, and almost every email service uses Naïve Bayes as a first-pass spam filter.

About this lab

AI-13 How AI and Machine Learning Work Classes 7-9 12 min Foundation

Learning objective: Test a simplified spam filter on sample messages and see which words push its decision.

What this simplifies: A small rule/keyword-weighted model is used; real spam filters combine many more signals.

Privacy: No learner input leaves the device.

Teacher prompt: Ask the class why this simulation might mislead someone who takes it too literally.

Reflect: What is one thing this activity showed you that you did not expect?

← Umbrella Decision Tree Back to All Labs Recommendation Engine →

Bring this learning to my school →