International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 05 | May 2025
p-ISSN: 2395-0072
www.irjet.net
Real-Time Bot Detection and Prevention on Websites Using Natural Interaction Analysis and Machine Learning J. Kalidass1, M. Mohamed Sadiq2, M. Vasanth Nanjappan3 1Assistant Professor, Department of CSE, Government College of Engineering Srirangam, Tamil Nadu, India 2,3UG Student, Department of CSE, Government College of Engineering Srirangam, Tamil Nadu, India
---------------------------------------------------------------------***-------------------------------------------------------------------
Abstract - Bots have become a significant threat to web
applications, performing automated tasks such as credential stuffing, data scraping, and click fraud. Traditional detection methods using CAPTCHAs and static fingerprinting are increasingly being bypassed by advanced bots that mimic human activity. This paper proposes a real-time bot detection and prevention system that leverages natural interactionbased behaviour such as mouse movements, keystroke dynamics, and navigation patterns to differentiate between human users and automated bots. The system captures these behaviours on the client side and analyses them on the server using a lightweight decision model. Experimental results demonstrate the effectiveness of the approach in identifying bots with high accuracy and minimal disruption to user experience. Key words: Web Security, Mouse Events, Key Strokes, Browser Fingerprinting, Natural Interaction Analysis, Bot Prevention.
1. INTRODUCTION The widespread integration of automation in web applications has led to a significant increase in botgenerated traffic. These bots pose serious threats to online platforms through activities such as data scraping, credential stuffing, fake account creation, and spamming. Traditional mitigation techniques-such as IP blocking, useragent filtering, and CAPTCHAs-are often ineffective against modern bots that can simulate human-like behaviour to bypass detection mechanisms. Recent advancements in browser automation frameworks have enabled bots to interact with websites more naturally, making it increasingly difficult to detect them using static rules or fingerprinting. Moreover, intrusive defences may degrade user experience and result in high false-positive rates, blocking legitimate users. Therefore, there is a growing demand for behaviour-driven, real-time bot detection mechanisms that are accurate, adaptive, and userfriendly.
© 2025, IRJET
|
Impact Factor value: 8.315
|
This paper presents a natural interaction-based bot detection and prevention framework, which captures realtime user behaviour data on the client side and communicates with a dedicated detection API for classification. The system monitors browser-level events such as mouse movement trajectories, keystroke patterns, scroll behaviour, and interaction timing. The captured data is sent via API to a backend detection engine that analyses the session using behavioural models and anomaly detection techniques. The primary advantage of this architecture is that the detection logic is decoupled from the client application, allowing for easy updates, scalability, and centralized monitoring. Suspicious sessions are flagged or blocked in real time, ensuring that bots are prevented from accessing critical application functions. The proposed approach balances security and usability by providing an accurate detection mechanism that minimizes false positives without burdening legitimate users. This system can be integrated seamlessly with various web platforms, offering a scalable and efficient solution for modern bot mitigation.
2. EXISTING SYSTEM Existing web bot detection systems predominantly utilize static fingerprinting, challenge-response mechanisms, or basic behavioural analysis. Although these methods provide a baseline defence against automated threats, they exhibit several critical limitations in the presence of modern, adaptive bots.
2.1 Static Fingerprinting-Based Detection Browser fingerprinting methods, such as those implemented in FP-Scanner [8] and FP-Radar [9], attempt to uniquely identify client devices based on attributes including screen resolution, installed plugins, and rendering behaviour. These static fingerprints are vulnerable to evasion techniques employed by advanced bots that randomize or
ISO 9001:2008 Certified Journal
|
Page 588