Context
This project focuses on interpretability and workflow clarity. Instead of using a transformer first, I built a classical baseline that is easy to debug, explain, and iterate.
Problem
The objective is binary classification from article text. The main challenge is balancing simplicity with useful predictive behavior while keeping training and inference lightweight.
Approach
I structured the pipeline so every stage is explicit:
- preprocessing for consistent text normalization,
- TF-IDF feature extraction,
- linear SVM training,
- evaluation and inspection.
This setup makes feature and model choices easy to change without rewriting the whole workflow.
Outcome
The project produces a stable baseline with clear diagnostics. It is useful both as a benchmark for future neural models and as a practical example of disciplined ML project structure.
Next Iteration
- Add calibrated probabilities for threshold-based decisions.
- Add a compact demo surface to test arbitrary user input quickly.