[Paper Review] BeCAPTCHA: Bot Detection in Smartphone Interaction using Touchscreen Biometrics and Mobile Sensors.

Alejandro Acien, Aythami Morales+3 more|arXiv (Cornell University)|May 27, 2020

User Authentication and Security Systems30 references10 citations

TL;DR

This paper proposes BeCAPTCHA, a novel CAPTCHA mechanism that leverages touchscreen interaction dynamics and accelerometer data to detect bots by modeling human behavioral patterns during a drag-and-drop task. Evaluated on the newly created HuMIdb dataset of 600 users across 14 sensors, the method demonstrates that mobile sensor data significantly enhances bot detection accuracy, especially when combined with GAN-synthesized adversarial samples.

ABSTRACT

In this paper we study the suitability of a new generation of CAPTCHA methods based on smartphone interactions. The heterogeneous flow of data generated during the interaction with the smartphones can be used to model human behaviour when interacting with the technology and improve bot detection algorithms. For this, we propose a CAPTCHA method based on the analysis of the information obtained during a single drag and drop task in combination with the accelerometer data. We evaluate the method by generating fake samples synthesized with Generative Adversarial Neural Networks and handcrafted methods. Our results suggest the potential of mobile sensors to characterize the human behaviour and develop a new generation of CAPTCHAs. The experiments are evaluated with HuMIdb (Human Mobile Interaction database), a novel multimodal mobile database collected for this work that comprises 14 mobile sensors acquired from 600 users. HuMIdb is freely available to the research community.

Motivation & Objective

• To develop a new generation of CAPTCHA systems that leverage smartphone sensor data to distinguish between human and bot interactions.
• To address the limitations of traditional CAPTCHAs by exploiting behavioral biometrics from mobile interactions.
• To evaluate the effectiveness of sensor-based behavioral modeling in detecting both synthetic and handcrafted bot attacks.
• To create and release a comprehensive multimodal mobile interaction database (HuMIdb) for future research in human behavior modeling and bot detection.

Proposed method

• The method analyzes touchscreen interaction patterns—such as pressure, speed, and trajectory—during a single drag-and-drop task on smartphones.
• It incorporates accelerometer data to capture device motion and user movement during interaction, enhancing behavioral fingerprinting.
• A machine learning model is trained to classify interactions as human or bot using features extracted from 14 mobile sensors.
• Synthetic bot samples are generated using Generative Adversarial Neural Networks (GANs) to simulate realistic adversarial behavior.
• The system is evaluated using a diverse set of attack vectors, including both GAN-generated and handcrafted bot interactions.
• The HuMIdb dataset, collected from 600 users, provides a rich, multimodal benchmark for training and testing the detection model.

Experimental results

Research questions

• RQ1 Can touchscreen interaction dynamics and accelerometer data effectively differentiate human users from automated bots in mobile applications?
• RQ2 How well does a sensor-based behavioral model perform against both GAN-synthesized and handcrafted bot attacks?
• RQ3 To what extent can multimodal sensor data improve the robustness and accuracy of CAPTCHA mechanisms compared to traditional text-based approaches?
• RQ4 How generalizable is the proposed detection model across diverse user populations and interaction patterns?

Key findings

• The integration of touchscreen and accelerometer data significantly improves bot detection performance compared to using either modality alone.
• The proposed method achieves high detection accuracy even against GAN-generated adversarial samples, demonstrating robustness to sophisticated synthetic attacks.
• The HuMIdb dataset, comprising 600 users and 14 synchronized sensor streams, provides a valuable benchmark for future research in mobile interaction analysis.
• The results indicate that behavioral biometrics from mobile interactions are a viable and effective foundation for next-generation CAPTCHA systems.
• The study confirms that mobile sensor data captures subtle human behavioral patterns that are difficult for bots to replicate.