Final year research project · Machine learning

Detect fake Instagram
accounts instantly

InstaGuard uses an XGBoost machine learning model to analyze behavioral patterns and profile metadata — classifying Instagram accounts as authentic or inauthentic with 99.2% accuracy.

99.2% Model accuracy
11 Features analyzed
5K Training samples
<1s Prediction time
InstaGuard Dashboard
Real account
99%
🛡️ Real account · 99% confidence
⚠️ Fake detected · High risk

Three steps to detection

InstaGuard analyzes account metadata patterns that strongly correlate with inauthentic behavior, then uses ML predictions to deliver a final verdict.

01

Input features

Enter 11 account features manually — including profile picture, follower counts, bio length, and username patterns — or upload a CSV for batch analysis.

02

ML processing

An XGBoost model trained on 5,000 samples analyzes 11 behavioral and structural features to compute a fake probability score for each account.

03

Verdict + explanation

Receive a real/fake verdict with confidence score, feature influence visualization, and a plain-English explanation of the prediction.

Everything you need

Single account analysis

Input 11 account features including profile picture, privacy status, and external URL — and receive an instant ML prediction with confidence gauge and feature influence visualization.

Real-time

Bulk file upload

Upload CSV or JSON files to analyze hundreds of accounts simultaneously with exportable results.

CSV · JSON

Model accuracy

XGBoost classifier trained with GridSearchCV hyperparameter tuning across 5-fold cross-validation, achieving 99.2% accuracy on the held-out test set.

GridSearchCV · 5-fold CV

Feature influence

See exactly which of the 11 features drove the prediction with animated influence bars and a plain-English explanation of why the model decided as it did.

Explainable AI

Demo presets

One-click demo accounts — bot, real, and edge case — to explore the model's behavior without manual entry.

Interactive
Model performance

Built on solid research

The XGBoost model was trained on 5,000 labeled Instagram profiles with stratified splitting, achieving near-perfect performance across all key classification metrics.

Read full methodology →
99.2%
Accuracy
99.6%
Precision
98.8%
Recall
99.2%
F1 score

Ready to detect fake accounts?

Try the live dashboard — no setup required.