Logistic RegressionNLPClassificationPython

SMS Spam Classification

Binary text classification with logistic regression — and learning what the metrics actually mean.

Role: Learning project
When: August 2026
Tools: Python, pandas, NumPy, scikit-learn, Matplotlib, Jupyter Notebook

Description

A binary text classifier that separates spam SMS messages from normal ones, built on the UCI SMS Spam Collection dataset.

TF-IDF turns each message into numbers, and logistic regression does the classifying. The more useful half of the project was everything after that: understanding why accuracy alone is a misleading score when one class is much rarer than the other.

What I worked through

TF-IDF feature extraction to turn SMS text into numerical features.
Logistic regression for binary classification.
Train/test splitting and evaluation.
Confusion matrices — and reading them properly.
Precision, recall and F1, and when each one is the score that matters.
Grid search for hyperparameter tuning, including on a larger search space.
Extending the same ideas to multi-class and multi-label problems.

Results

A confusion matrix: 1,206 ham correctly kept, no ham wrongly binned, 148 spam caught and 39 spam missed. — The held-out test set. 1,206 ordinary messages kept and not one of them wrongly binned; of the spam, 148 caught and 39 let through. Erring that way round is the right way round for a spam filter — a lost real message costs more than one that slips into the inbox.

A ROC curve rising almost immediately to the top-left corner, with AUC marked as 0.99. — The same classifier as a ROC curve — AUC 0.99.