Article
Machine Learning-Based Detection of Fake Profiles on Social Media Platforms
The widespread adoption of social media platforms has given rise to a parallel epidemic of fraudulent accounts designed to manipulate engagement metrics, distribute misinformation, and conduct phishing operations at scale. This paper proposes a machine learning-based system for automatically detecting fake Instagram profiles by analyzing behavioral and profile-level features extracted from labeled account datasets. The proposed framework processes structured attributes—including follower and following counts, posting frequency, engagement ratios, and profile completeness indicators—through a preprocessing pipeline that handles missing values, normalizes numerical features, and engineers derived attributes such as the follower-to-following ratio. Classification is performed using a Random Forest ensemble model and a multi-layer Neural Network trained with the Adam optimizer and categorical cross-entropy loss. Model evaluation on held-out test data demonstrates classification accuracy exceeding 90%, with strong precision and recall scores across both genuine and fraudulent classes. The system further integrates an OpenCV-based face detection module and an EasyOCR text extraction pipeline to analyze profile images, enabling richer feature construction. A web-based dashboard provides real-time prediction, visualization of results, and an intuitive interface for analysts. Results confirm that the combined use of behavioral feature engineering and neural classification substantially outperforms traditional rule-based detection mechanisms in identifying sophisticated fake accounts.
Full Text Attachment
