Dynamic time warping como métrica de similaridade entre os gestos de usuários de smartphones com finalidade de validação de identidade

Abstract

This work aims to evaluate the impact of metrics generated with the help of the Dynamique Time Warping algorithm on anomaly detection models trained to authenticate a mobile user through inherited factor based on their interaction with a smartphone screen. The research was conducted through data collection from 33 users who agreed to interact with a mobile application developed for this work, aiming to maximize the number of swipe gestures performed by the user during the collection process. From the collected data, two Isolation Forest models were trained: one reference model that did not use Dynamique Time Warping metrics and another that did. The results indicated an absolute improvement of 3.725%, which represents a relative improvement of 8.40% in the model’s task of classifying intruder gestures as anomalous. This work contributes to the field of device and mobile application security, as it proposes an enhancement to an authentication method focused on smartphones. Consequently, it also contributes to the security of smartphone users, who make up over 68% of the global population.