Spam Email Detection — Classical ML Benchmark

01Overview

A study of how far well-understood classical models get on spam classification, using the UCI Spambase dataset. The point was not to reach for the largest available model but to compare a set of standard approaches carefully and report what actually held up.

02Method and result

Naive Bayes, SVM, Random Forest, and a voting ensemble were trained and evaluated under the same conditions, with results reported on a held-out test set rather than on training data.

Random Forest came out ahead at 94.7% accuracy and 93.1% F1. The ensemble did not beat the strongest single model, which is a useful result in itself: stacking models is not automatically an improvement.