From Logistic Regression to GPT-2: Building a Complete Spam Detection & Sentiment Analysis Pipeline

Benchmarking eight models across three paradigms revealing why accuracy is a deceptive metric for imbalanced text classification Table of Contents The Dataset & Class Imbalance What Spam Looks Like in Words Phase 1: The Eight-Model Benchmark How All Eight Models Stack Up Where Each Model Actually Fails Discrimination Ability Across Thresholds The Honest Test Under Imbalance Phase 2: Sentiment Enrichment Key Lessons What’s Next Every day, an estimated 3.4B phishing emails are sent worldwide. Spam detection is one of the oldest problems in machine learning and one of the most instructive.