Machine Learning-Based Phishing Email Detection System: TF-IDF and Naive Bayes Achieve 97.82% Accuracy

This article introduces a phishing email detection system using TF-IDF text vectorization and Naive Bayes classifier, which achieves a classification accuracy of 97.82% on the test dataset and supports real-time email prediction functionality.

• Original Author/Maintainer: Anurag Upadhyay
• Source Platform: GitHub
• Original Title: Phishing-Email-Detector
• Original Link: https://github.com/anurag21112006/Phishing-Email-Detector
• Publication Date: June 9, 2026

In the digital age, email remains the primary vector for phishing attacks. Phishing emails not only threaten the privacy and security of individual users but also serve as the main entry point for corporate data breaches. Statistics show that over 90% of cyberattacks start with phishing emails. Traditional rule-based filtering methods struggle to cope with evolving phishing techniques, so using machine learning to automatically identify phishing emails has become an important research direction in the cybersecurity field.

This project is a machine learning-based phishing email detection system that can automatically classify emails into "safe emails" or "phishing emails". The system uses Natural Language Processing (NLP) technology combined with a Naive Bayes classifier to identify potential malicious emails by analyzing content features of the emails.

• Automatic Email Classification: Label emails as safe or phishing category
• TF-IDF Text Vectorization: Convert text into numerical feature vectors
• Naive Bayes Machine Learning Model: Efficient probabilistic classification algorithm
• Accuracy Evaluation: Quantitative metrics for model performance
• Confusion Matrix Visualization: Intuitively display classification results
• Real-time Email Prediction: Support instant detection of new emails
• Model Persistence: Save trained models using Pickle

The project uses a classic combination from the Python ecosystem:

• Python: Core programming language
• Pandas: Data processing and cleaning
• Scikit-Learn: Machine learning algorithm implementation
• Matplotlib: Visualization chart generation
• Pickle: Model serialization and deserialization

The system uses a dataset containing email text and corresponding labels for training:

Field	Description
text_combined	Email body content
label	Classification label (0 = safe email, 1 = phishing email)

The entire detection process follows a standard machine learning workflow:

1. Data Loading: Read the email dataset from CSV files
2. Text Preprocessing: Clean and standardize email text content
3. Feature Engineering: Convert text to numerical features using TF-IDF
4. Data Splitting: Split the dataset into training and test sets
5. Model Training: Train the classifier using Naive Bayes algorithm
6. Performance Evaluation: Calculate accuracy and generate confusion matrix
7. Real-time Prediction: Classify new emails