ML Based Solution to Refine Captcha

ML Based Solution to Refine Captcha

Authors: Tatineni Leela Phanindra, Murari Sai Deepak, Bandi Shashidhar Reddu

Affiliation: Computer Science and Engineering Contact: tatineniphanindra@gmail.com

Date: May 2025

Abstract
Traditional CAPTCHA mechanisms, though widely used for securing web portals, are increasingly criticized for poor user experience and limited accessibility. In response, the Unique Identification Authority of India (UIDAI) seeks to replace CAPTCHA with a passive alternative that can seamlessly differentiate between bots and legitimate users without explicit user intervention. This research presents a novel passive CAPTCHA solution that leverages browser-based environmental parameters and interaction behaviors. A JavaScript-based frontend captures user-specific data like

mouse entropy, screen resolution, and interaction timing. A backend machine learning model, trained using Random Forest, classifies the session as either human or bot in real-time. The solution achieves 96.2% accuracy and minimal response latency (<200ms), significantly outperforming traditional CAPTCHA systems in both usability and performance. This paper discusses the methodology, system architecture, model evaluation, and real-world testing, along with ethical considerations and future enhancements for adaptive security.

Keywords
Passive CAPTCHA, UIDAI, Browser Context, Bot Detection, Machine Learning, Web Security, API Protection, User Experience.