About the project
Fake KYC messages, lottery prizes, job offers and "parcel held" SMS cheat thousands of mobile users every day. Spam SMS and Phishing URL Detection (app name SpamKavach) is an Android app that checks a message or a link *before* the user acts on it. The check runs fully on the phone using two machine learning models, so it is instant, private and works even without internet. Users can report scams, moderators confirm them, and every phone receives the updated blacklist automatically.
How it works
- The user shares an SMS from any messaging app, or pastes a message or link on the home screen.
- A Multinomial Naive Bayes model (pure Java, inside the app) gives the spam percentage and highlights the words that made the message suspicious.
- Every link is split into 17 features (IP address, risky domain ending, brand word on a fake site, short link, hyphens, https and more) and scored by a Logistic Regression phishing model.
- Links and numbers are matched with a community blacklist saved on the phone; official SMS headers (like HDFCBK) and official domains are treated as safe.
- The result is shown as SPAM / SUSPICIOUS / SAFE with scam type, reasons and a safety tip, and can be saved, reported or shared as a warning.
- Moderators review reports; a confirmed scam is added to the blacklist and all phones sync the new version.
Users
- User: checks SMS and links, keeps history, reports scams, reads city scam alerts, takes a safety quiz and downloads a PDF safety report.
- Moderator: reviews the report queue with model score and same-sender count, manages the blacklist and posts scam alerts.
- Admin (web panel): manages users, moderators, categories and keywords, trusted senders, quiz bank, detection levels and printable reports.
Highlights
- 3 roles, 25 Android screens, 14 admin pages, 29 PHP APIs, 14 MySQL tables
- Two ML models with real test results: SMS Naive Bayes 98.39% accuracy, URL model 98.90% accuracy
- Python training script included (re-train with one command), plus datasets and confusion matrices
- Share-to-app, clipboard detection, offline checking, version-based blacklist sync
- Scam alerts, safety quiz, badges, daily streak, Kavach score and A4 PDF safety report
- Validation on every form in the app and in PHP, bcrypt passwords, token based API security
- Demo data and demo logins for all roles
What you get
- Complete Android Studio project (Java) and PHP admin panel + APIs
- MySQL database file with sample data
- ML training script, datasets and accuracy results
- Project report (Word, 55 pages) with all UML diagrams
- Project abstract (2 pages), project explanation with 20 viva questions, setup guide PDF
- Help with setup on your laptop
Who can use this project
Suitable as a final year project for BE / B.Tech (Computer, IT, AI & DS), MCA, BCA, M.Tech and Diploma students. A good choice if you need an Android project with source code that has real machine learning, a cyber security theme, an admin panel and full documentation.
Frequently asked questions
Does the app read my SMS inbox? No. The user shares or pastes a message, so no SMS permission is needed and privacy is kept. Is internet needed to check a message? No. Both models run inside the phone; internet is needed only for history, reports and blacklist sync. Can I train the model on my own data? Yes. Put your CSV in the dataset folder and run train_model.py; the new model files are written into the app automatically. Which ML algorithms are used? Multinomial Naive Bayes for SMS text and Logistic Regression on 17 URL features for links.
