About the project
Fake forwards spread faster than corrections, and most people have no quick way to check a message before sharing it. This fake news detection Android project (app name: TruthLens) solves that with a machine learning model that runs fully on the phone. The reader pastes or shares a message and gets REAL, FAKE or UNSURE in milliseconds. The app highlights the words that pushed the result and shows clickbait and source signals. When the model is wrong, human fact-checkers correct it, and their corrections become new training data.
How it works
- The reader pastes news text, shares it from WhatsApp, or enters a link (the app reads the page headline)
- The text is cleaned and turned into TF-IDF values for 3,000 words and word pairs, plus 3 style values (CAPS, !!!, links)
- A neural network in TensorFlow Lite (3,003 → 32 → 2) gives the fake chance on the phone, with no internet needed
- A trusted / unreliable / satire source list from the admin adjusts the score; 60+ is FAKE, below 40 is REAL
- Words that pushed the result are highlighted red or green, and a matching verified fact-check is shown
- Wrong results are reported with a reason and a screenshot; fact-checkers review grouped claims in a priority queue
- Accepted corrections are exported as a training CSV for the next model version
Users
- Reader: checks news, sees explained results, reports wrong ones, reads fact-checks, plays the daily quiz, earns badges
- Fact-checker: approval-based login, reviews reported claims, gives verdicts with sources, publishes fact-checks
- Admin (web panel): dashboard charts, user approval, categories, source list, reports, quiz, notices, ML settings
Highlights
- 3 roles, 24 Android screens, 14 admin pages, 29 PHP APIs, 13 MySQL tables
- Machine learning project with real test results: 97.12% accuracy, 0.971 F1 score on 800 test lines
- Works offline: TensorFlow Lite model plus a plain-Java backup of the same network
- Explainable result (word highlighting), source credibility check, link reading, WhatsApp share support
- Spot-the-Fake daily quiz, points, streaks, badges, leaderboard and a weekly PDF report
- Validations in both the app and the PHP side, bcrypt passwords, PDO prepared statements, CSRF tokens in the admin panel
- Sample data with demo logins for reader, fact-checker and admin; charts show recent data right after import
What you get
- Complete Android Studio source code (Java) and the PHP admin panel + APIs
- MySQL database file with sample data
- Python training script, dataset and the exported model files
- Project report (Word, 54 pages) with all UML diagrams, plus a 2-page abstract (Word + PDF)
- Project explanation with 22 viva questions and answers
- Setup guide (Word + PDF), README and screenshots, with setup help
Who can use this project
Good for BE / B.Tech, MCA, BCA and Diploma students of CSE, IT and ENTC who want a final year project in machine learning on Android with source code and documentation. It also works as an NLP text classification project, a mini project with an admin panel, or an Android ML project for a seminar or demo.
Frequently asked questions
Does the fake news check need internet? No. The model runs on the phone with TensorFlow Lite. Internet (Wi-Fi to your PC) is needed only to save history, reports and fact-checks.
Can I retrain the model with my own data? Yes. Add lines to the CSV, run train_model.py and copy the 4 output files into the app's assets folder.
Which algorithm is used and why not BERT? TF-IDF features with a small neural network. It is 385 KB, fast and offline, while BERT needs hundreds of MB and a server.
How is a wrong prediction corrected? Readers report it, a fact-checker gives the final verdict, and accepted cases are exported from the admin panel as training data.
