📰 Classifying BBC News Articles with Machine Learning

In this project, I explored how machine learning can help categorize BBC news articles into topics like business, politics, sport, entertainment, and tech. The idea was to see how well different models could understand the content of an article and assign it to the right category—without any human input.

Getting Started

The dataset comes from a Kaggle competition and includes a mix of labeled and unlabeled articles. Before diving into modeling, I spent some time cleaning the data—removing duplicates, checking for balance across categories, and making sure everything was in the right format.

Preprocessing the Text

To prepare the articles for analysis, I used a technique called TF-IDF, which helps highlight the most meaningful words in each article. I also tried a few tweaks, like replacing numbers with a placeholder, to see if that would improve performance. Turns out, it didn’t help much—some numbers (like years) actually carry useful context.

Unsupervised Learning: Finding Patterns Without Labels

I started with an unsupervised approach using Non-negative Matrix Factorization (NMF). This method doesn’t rely on labeled data—it just looks for patterns in the text. Surprisingly, it did quite well, reaching around 91% accuracy after some tuning.

Supervised Learning: Training with Labels

Next, I tried supervised models, which learn from labeled examples. I used Logistic Regression and LinearSVC, and both performed even better than NMF. With enough training data, they reached up to 97% accuracy.

What stood out was how efficient LinearSVC was—it managed to get solid results even with a smaller portion of the training data.

Final Thoughts

This project was a great way to compare different approaches to text classification. It showed that while unsupervised models can be useful, supervised learning tends to be more accurate when labels are available. It also highlighted how preprocessing choices can impact performance in subtle ways.

If you're curious about the details, the full notebook is embedded below 👇

You can also view the notebook in a separate page, or check it on GitHub.