eXplainable AI(XAI) for Touch-Stroke Biometrics: Insights from SHAP

Soodamani Ramalingam; Dominic Lovric; Ooi Shih  Yin; Richard Guest; Moises  Diaz; Fabio Garcia; David Lawunmi

eXplainable AI(XAI) for Touch-Stroke Biometrics: Insights from SHAP

Soodamani Ramalingam, Dominic Lovric, Ooi Shih Yin, Richard Guest, Moises Diaz, Fabio Garcia, David Lawunmi

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

This paper presents an XAI-based framework for touch-stroke behavioural biometrics. Initially, a Random Forest classifier is trained to perform user classification, and feature importances are derived from the model's internal metrics. Subsequently, SHAP explanations are applied to obtain model-agnostic feature attributions, in both portrait and landscape modes. A comparison between the two approaches is then conducted to identify consistent patterns of feature relevance, informing the decision to exclude redundant or less influential features. The findings underscore the potential of integrating XAI into behavioural biometrics to enhance transparency and user trust.

Original language	English
Title of host publication	The 57th Annual ICCST
Subtitle of host publication	Proceedings
Publisher	IEEE Xplore Digital Library
Pages	221
Number of pages	226
Publication status	Accepted/In press - 15 Oct 2025
Event	IEEE International carnaham Conference on Security Technology - University of Texas at San Antonio (UTSA) School of Data Science, San Antonio, Texas, Texas, United States Duration: 14 Oct 2025 → 17 Oct 2025 https://site.ieee.org/iccst/2025-san-antonio-texas-usa/

Conference

Conference	IEEE International carnaham Conference on Security Technology
Abbreviated title	2025 ICCST
Country/Territory	United States
City	Texas
Period	14/10/25 → 17/10/25
Internet address	https://site.ieee.org/iccst/2025-san-antonio-texas-usa/

Keywords

Explainable AI (XAI)
touch-stroke dynamics
biometrics
SHAP

Embargoed Document

Accepted_Version
Accepted author manuscript, 786 KB
Licence: Unspecified

Cite this

@inproceedings{897b2bd4be3e49b2bfbd78724ee69e55,

title = "eXplainable AI(XAI) for Touch-Stroke Biometrics: Insights from SHAP",

abstract = "This paper presents an XAI-based framework for touch-stroke behavioural biometrics. Initially, a Random Forest classifier is trained to perform user classification, and feature importances are derived from the model's internal metrics. Subsequently, SHAP explanations are applied to obtain model-agnostic feature attributions, in both portrait and landscape modes. A comparison between the two approaches is then conducted to identify consistent patterns of feature relevance, informing the decision to exclude redundant or less influential features. The findings underscore the potential of integrating XAI into behavioural biometrics to enhance transparency and user trust.",

keywords = "Explainable AI (XAI), touch-stroke dynamics, biometrics, SHAP",

author = "Soodamani Ramalingam and Dominic Lovric and Yin, \{Ooi Shih\} and Richard Guest and Moises Diaz and Fabio Garcia and David Lawunmi",

year = "2025",

month = oct,

day = "15",

language = "English",

pages = "221",

booktitle = "The 57th Annual ICCST",

publisher = "IEEE Xplore Digital Library",

url = "https://site.ieee.org/iccst/2025-san-antonio-texas-usa/",

}

TY - GEN

T1 - eXplainable AI(XAI) for Touch-Stroke Biometrics: Insights from SHAP

AU - Ramalingam, Soodamani

AU - Lovric, Dominic

AU - Yin, Ooi Shih

AU - Guest, Richard

AU - Diaz, Moises

AU - Garcia, Fabio

AU - Lawunmi, David

PY - 2025/10/15

Y1 - 2025/10/15

N2 - This paper presents an XAI-based framework for touch-stroke behavioural biometrics. Initially, a Random Forest classifier is trained to perform user classification, and feature importances are derived from the model's internal metrics. Subsequently, SHAP explanations are applied to obtain model-agnostic feature attributions, in both portrait and landscape modes. A comparison between the two approaches is then conducted to identify consistent patterns of feature relevance, informing the decision to exclude redundant or less influential features. The findings underscore the potential of integrating XAI into behavioural biometrics to enhance transparency and user trust.

AB - This paper presents an XAI-based framework for touch-stroke behavioural biometrics. Initially, a Random Forest classifier is trained to perform user classification, and feature importances are derived from the model's internal metrics. Subsequently, SHAP explanations are applied to obtain model-agnostic feature attributions, in both portrait and landscape modes. A comparison between the two approaches is then conducted to identify consistent patterns of feature relevance, informing the decision to exclude redundant or less influential features. The findings underscore the potential of integrating XAI into behavioural biometrics to enhance transparency and user trust.

KW - Explainable AI (XAI)

KW - touch-stroke dynamics

KW - biometrics

KW - SHAP

M3 - Conference contribution

SP - 221

BT - The 57th Annual ICCST

PB - IEEE Xplore Digital Library

T2 - IEEE International carnaham Conference on Security Technology

Y2 - 14 October 2025 through 17 October 2025

ER -

eXplainable AI(XAI) for Touch-Stroke Biometrics: Insights from SHAP

Abstract

Conference

Keywords

Embargoed Document

Fingerprint

Cite this