Hybrid Machine Learning-Based Instagram Fake Account Detection with Behavioral Scoring

Abstract

Fake accounts on Instagram are a real problem — they spread false information, run scams, impersonate real people, and quietly shift public opinion for whoever's behind them. What makes them hard to catch is that well-built ones actually look human: reasonable follower counts, consistent posting, normal usernames. Flagging them manually doesn't scale. The system this paper describes automates the process through three layers — pulling live profile data via instaloader, checking against a stored dataset of known users, and running a Random Forest classifier on anything new. The features going into that model are practical ones: follower-to-following ratio, posting frequency, bio presence, external links, verification status, and general activity patterns. The core finding is that combining those signals outperforms any single-feature or single-model approach — not a dramatic result, but a useful one. More importantly, the system is built to actually be deployed, not just benchmarked in a paper