Touch-Based Hybrid Authentication with Accurate Feature Extraction and Hyperparameter Tuning for Biometric Applications
T P Ramachandran
Abstract
Touch-Based Hybrid Authentication uses behavioral biometrics along with machine learning to authenticate a user on the basis of their one-of-a-kind touchscreen behaviors, for example, taps, swipes, and pattern locks. The Touch-based Hybrid Authentication Technology with Hybrid Biometric (THAFHB) is a system for user verification that integrates biometrics with Deep Learning (DL). The two Android applications on a Samsung Galaxy SII take the touch gesture data. The applications also measure user ID, type of gesture, pressure, and timestamps. The framework employs a modified DarkNet-53 architecture that does have Res Net connections to improve feature learning. To improve performance and calibrate the parameters, the framework utilizes a recently developed algorithm inspired by rabbit behavior, the Adaptive Rabbit Optimization (ARO) algorithm, to facilitate a trade-off between exploration and exploitation. A Hyper Parameter Optimization (HPO) algorithm, based on predator-prey dynamics, is used to set the parameters for ShuffleNetv2.3. It follows a two-phase process, i.e. exploration and exploitation. The result discussion by mean squared error calculation, validation rate vs noise level calculation, mean absolute error vs noise level calculation, THAFHB overall performance calculation, and THAFHB confusion matrix calculation.