Implementation Of Long Short-Term Memory (LSTM) For User Authentication Based On Keystroke Dynamics

Admir Ferhatovic,Ali Abd Almisreb,Sherzod Turaev,Mohammed A Saleh

doi:10.21533/scjournal.v9i1.187

Admir Ferhatovic, Ali Abd Almisreb + Show 2 more

Open Access

https://doi.org/10.21533/scjournal.v9i1.187

Copy DOI

Abstract

The paper reviews the usage, development and implementation of keystroke dynamics as a viable authentication system. AI methods are advancing, but we are still lacking biometric authentication systems in modern software which is being used daily. This paper shows the usage of long short-term memory layers for solving problems like keystroke dynamics and efficiently shows that with modern hardware, training and maintaining a small model is not taxing on the resources, as it may have been.

Highlights

INTRODUCTIONIn this paper I want to provide a clear overview of a keystroke dynamics classifier implementation using longshort-term-memory layers and present an idea about how many test samples per person are needed for a secure and maintainable system [1]
In this paper I want to provide a clear overview of a keystroke dynamics classifier implementation using longshort-term-memory layers and present an idea about how many test samples per person are needed for a secure and maintainable system [1]. This way I want to encourage the usage of keystroke dynamics as an authentication system in everyday applications [2], not to replace the simple password system, but to act as a support for it, this way maintaining a high level of security without needing the user to take additional actions which distract him from his everyday tasks [3]
Whether an authentication system should use only keystroke dynamics on its own, will be discussed later in this paper, after we see the potential accuracy of such a system

Summary

INTRODUCTION

In this paper I want to provide a clear overview of a keystroke dynamics classifier implementation using longshort-term-memory layers and present an idea about how many test samples per person are needed for a secure and maintainable system [1]. Having users typing rhythms would help detect if an account is being shared and help recognize if a policy is being broken. For this purpose, a publicly available dataset was used which contains 400 samples of time series vectors for 51 users [8]

LITERATURE REVIEW

METHODS

RESULTS

CONCLUSION