AI / ML Project

Fake News Detection System

A text classification system that scores news articles for credibility and explains which words influenced the decision.

Intermediate7–9 weeksCSE · IT · MCAAI & Machine LearningCybersecurity

Project overview

The system classifies a news article or headline as likely reliable or likely fabricated, using linguistic patterns learned from a labelled corpus. Alongside the label it shows the terms that pushed the classifier towards its decision, which turns an opaque prediction into something a user can evaluate.

The project covers a complete NLP pipeline — cleaning, vectorisation, model comparison and explanation — and is a strong choice for teams who want a language project without needing a GPU.

Problem statement

Fabricated news spreads fastest on messaging platforms where there is no source attribution at all. Manual fact checking cannot keep pace, and most automated tools give a verdict with no reasoning, which does little to build the reader's own judgement.

Proposed solution

Train text classifiers on a labelled news corpus using TF-IDF and transformer embeddings, compare their performance honestly, and deploy the better model behind a web interface. Add an explanation layer that surfaces the highest-weighted terms so the user sees why the article was flagged.

Key features

  • Paste text or submit a URL for analysis
  • Text cleaning pipeline: lowercasing, stopword removal, lemmatisation
  • Comparison of at least three classifiers with a documented results table
  • Credibility score with a confidence value
  • Explanation panel highlighting the most influential terms
  • Batch analysis mode for evaluating a set of articles
  • Analysis history stored per user

Technology stack

NLP

PythonNLTKspaCyTF-IDFHugging Face Transformers

Models

Logistic RegressionNaive BayesLinear SVMDistilBERT (optional)

Backend

FlaskSQLite

Front end

ReactBootstrap or Tailwind CSS

System architecture

The flow below is what goes into your architecture diagram and the corresponding chapter of the report.

  1. Input handler accepts raw text or fetches and extracts article body from a URL
  2. Preprocessing normalises the text and removes noise
  3. Vectoriser converts text to TF-IDF features or transformer embeddings
  4. Classifier returns a label with a probability
  5. Explainer extracts the top contributing features for that prediction
  6. Web layer displays label, confidence and explanation together

Expected output

  • Credibility label with probability
  • Highlighted influential terms for the article
  • Model comparison table across the classifiers tried
  • Confusion matrix and per-class precision and recall

Suitable branches, difficulty and duration

Suitable branches

Difficulty level

Intermediate. Comfortable for a team where at least one member has worked in this stack before.

Development duration

7–9 weeks including documentation, tested and ready to demonstrate. Milestones are planned around your review dates.

Documentation, review, demo and viva support

Documentation support

Abstract, problem statement, literature survey guidance, system architecture, ER and UML diagrams, module descriptions, test cases and screenshots — prepared in your department format.

Project review support

Delivery is split across your review calendar so review one has a design and plan, review two has working modules, and the final review has the integrated, tested system.

Demo support

A rehearsed demo path, seeded sample data, a reset script, and a recorded walkthrough kept as a backup in case something fails in the review room.

Viva preparation

A written question bank for this domain, a code walkthrough session, and practice at explaining each module in your own words — because the examiner will ask you.

Customisation options

Colleges reject duplicate topics, so most projects are adjusted before we start. These are the usual directions for this one.

  • Add regional language support for Tamil or Hindi news with a suitable corpus
  • Add a source reputation feature based on domain history
  • Extend to a browser extension that scores pages as they are opened
  • Add clickbait headline detection as a separate classifier
Talk to CodeLab69

Enquire about the Fake News Detection System project

Send your branch, college and first review date. You will get confirmation of scope, what the customisation would look like, and a delivery plan mapped to your reviews.