This project implements a real-world spam email classifier using the Spambase dataset. It applies Naive Bayes classification with hyperparameter tuning, feature selection, bootstrap confidence intervals, and includes model comparison with Logistic Regression.
- Feature pruning (low variance & high correlation)
- Hyperparameter tuning (kernel + width)
- Stratified cross-validation
- Evaluation metrics: Accuracy, Precision, Recall, F1, AUC
- Bootstrap Confidence Interval
- ROC and Precision-Recall curves
- Model comparison with Logistic Regression
- MATLAB deployable model (
SpamClassifierModel.mat)
- Name: Spambase Dataset from UCI
- Size: 4,601 emails (57 features + 1 target)
- Target Classes: Spam (1), Non-Spam (0)
naive-bayes-spam-detection/
├── NaiveBayes_SpamDetection.m # Main MATLAB script
├── spambase.csv # Dataset file (not included in public repo)
├── outputs/ # Output visualizations
│ ├── class_distribution.png
│ ├── hyperparameter_eval.png
│ ├── optimization_surface.png
│ ├── confusion_matrix.png
│ ├── roc_curve.png
│ └── precision_recall_curve.png
├── SpamClassifierModel.mat # Saved deployable model
└── README.md # Project documentation
| Metric | Value |
|---|---|
| Cross-Validated Accuracy (Tuning) | 85.07% |
| 5-Fold CV Accuracy | 84.85% |
| Test Set Accuracy | 85.94% |
| Precision | 0.83752 |
| Recall | 0.79742 |
| F1-Score | 0.81698 |
| AUC | 0.91582 |
| Bootstrap CI (95%) | [83.99%, 87.61%] |
| Logistic Regression Accuracy | 87.68% |
- Download the dataset from UCI Spambase
- Open
NaiveBayes_SpamDetection.min MATLAB - Run the script — metrics and figures will be generated automatically
A trained model is saved using MATLAB's saveLearnerForCoder for production deployment:
saveLearnerForCoder(finalModel, 'SpamClassifierModel')
Also available on Kagggle: Kaggle
Victor Collins Oppon
MSc Data Science | Chartered Accountant
LinkedIn Profile
Building a bridge between finance and intelligent automation.
This project is licensed under the MIT License.
You are free to use, modify, and distribute this software for both personal and commercial use, provided proper attribution is given.