[Paper Review] Recurrent neural network-based user authentication for freely typed keystroke data

Junhong Kim, Pilsung Kang|arXiv (Cornell University)|Jun 16, 2018

User Authentication and Security Systems15 references8 citations

TL;DR

This paper proposes an RNN-based keystroke dynamics authentication (RNN-KDA) model that processes variable-length, freely typed text by using consecutive key pairs as input and their actual typing times as targets. It achieves a 5-6% equal error rate (EER) with just 10 test keystrokes—significantly outperforming benchmark methods, which exceed 20% EER, and showing faster, more stable performance gains with increased training length.

ABSTRACT

Keystroke dynamics-based user authentication (KDA) based on long and freely typed text is an enhanced user authentication method that can not only identify the validity of current users during login but also continuously monitors the consistency of typing behavior after the login process. Previous long and freely typed text-based KDA methods had difficulty incorporating the key sequence information and handling variable lengths of keystrokes, which in turn resulted in lower authentication performance compared to KDA methods based on short and fixed-length text. To overcome these limitations, we propose a recurrent neural network (RNN)-based KDA model. As the RNN model can process an arbitrary length of input and target sequences, our proposed model takes two consecutive keys as the input sequence and actual typing time for the corresponding key sequence as the target sequence. Based on experimental results involving 120 participants, our proposed RNN-KDA model yielded the best authentication performance for all training and test length combinations in terms of equal error rate (EER). It achieved a 5%-6% EER using only 10 test keystrokes while the EERs of other benchmark methods were above 20%. In addition, its performance steadily and more rapidly improves compared to the benchmark methods when the length of training keystrokes increases.

Motivation & Objective

• To address the limitations of existing keystroke dynamics authentication (KDA) methods that struggle with variable-length, freely typed text and poor sequence modeling.
• To improve authentication accuracy by incorporating key sequence information and handling variable input lengths in long text.
• To develop a robust, end-to-end RNN-based model that learns temporal typing patterns from real-world, unconstrained typing behavior.
• To evaluate the model’s performance across varying training and test sequence lengths, especially with minimal data.

Proposed method

• The RNN-KDA model uses two consecutive keystrokes as input sequences to capture sequential typing patterns.
• It predicts the actual typing time (inter-keystroke interval) for each key pair, using this as the target sequence.
• The model leverages the RNN’s ability to process variable-length sequences, enabling effective handling of diverse typing lengths.
• Training involves learning the mapping from key pairs to their corresponding timing behavior, capturing user-specific typing dynamics.
• The architecture is trained end-to-end on keystroke data from 120 participants, optimizing for authentication accuracy.
• The model is evaluated using equal error rate (EER) across multiple training and test length combinations.

Experimental results

Research questions

• RQ1 Can an RNN-based model effectively model variable-length, freely typed keystroke sequences for user authentication?
• RQ2 How does the proposed RNN-KDA model compare to existing KDA methods in terms of authentication accuracy, especially with short test sequences?
• RQ3 Does the performance of the RNN-KDA model improve more rapidly and steadily with increasing training sequence length compared to benchmark methods?
• RQ4 To what extent does incorporating key sequence information enhance authentication performance in unconstrained typing scenarios?

Key findings

• The RNN-KDA model achieved a 5-6% equal error rate (EER) using only 10 test keystrokes, significantly outperforming benchmark methods that exceeded 20% EER.
• The model’s authentication performance improved steadily and more rapidly than benchmark methods as the length of training keystrokes increased.
• The RNN-KDA model demonstrated superior generalization across all training and test length combinations evaluated.
• The model effectively captured sequential typing patterns by modeling consecutive key pairs and their timing, enhancing user-specific behavioral modeling.