Final Year Project · Islamia University of Bahawalpur
Check whether a news story holds up.
Paste any English news article and the system runs it through an NLP pipeline and a trained Naive Bayes classifier, then returns a Real or Fake verdict with a confidence score and the words that drove the decision.
95.45%
Model accuracy
110
Training articles
840
Vocabulary size
10
Articles analysed
How an article is processed
Six stages, end to end01
Clean
Lowercase, strip punctuation, numbers and URLs
02
Tokenise
Split the text into individual word tokens
03
Remove stop words
Drop 180+ common English filler words
04
Stem
Reduce each word to its root form
05
Extract features
Bag of Words weighted with TF-IDF
06
Classify
Multinomial Naive Bayes returns the verdict
Current model
| Algorithm | Multinomial Naive Bayes |
|---|---|
| Features | Bag of Words + TF-IDF |
| Accuracy | 95.45% |
| Precision | 91.67% |
| Recall | 100.00% |
| F1 score | 95.65% |
| Train / test split | 88 / 22 articles |
What the system does not do
Scope is deliberately limited to what the project can support honestly:
- English-language text only — no Urdu or other languages
- Text only — images, video and audio are out of scope
- No source or author credibility checks against external databases
- No live scraping of social media feeds
- Satire and opinion pieces are not distinguished from fabricated news
The verdict is a statistical signal to support a reader's judgement, not a final ruling on any story.