Investigation Of A Method To Continuously Authenticate Users With Mouse Movements

Machine Learning 20/05/2024

3 main points
✔️ focus on the importance of efficient and reliable user authentication methods in the computer security field
✔️ study results highlight the versatility of the model used, with PC mouse dynamics providing a viable method for continuous authentication
✔️mouse movement dynamics corroborated as a useful tool for continuous user authentication

From Clicks to Security: Investigating Continuous Authentication via Mouse Dynamics
written by Rushit Dave, Marcho Handoko, Ali Rashid, Cole Schoenbauer
(Submitted on 6 Mar 2024)
Comments: Published on arxiv.
Subjects: Artificial Intelligence (cs.AI)

code：

The images used in this article are from the paper, the introductory slides, or were created based on them.

Summary

This paper focuses on the importance of efficient and reliable user authentication methods in the field of computer security. It examines the possibility of using mouse movement dynamics as a new approach for continuous authentication. Specifically, we analyzed user mouse movements in two different game scenarios, "Team Fortress" and "Poly Bridge," to investigate behavioral patterns specific to high- and low-intensity UI interactions.

In this study, we employed machine learning models to go beyond traditional methodologies and evaluate their effectiveness in capturing subtle behaviors reflected in mouse movements. This approach allowed for a more nuanced and comprehensive understanding of user interaction patterns.

Study results suggest that mouse movement characteristics may serve as a reliable indicator of ongoing user verification. In addition, the machine learning model used in this study performed better in user verification than traditional methods.

Ultimately, this research contributes to the development of enhanced computer security and robust authentication systems, and demonstrates the potential for leveraging user behavior, particularly mouse dynamics.

Introduction

The paper points out that in the rapidly evolving cybersecurity landscape, traditional authentication methods are vulnerable to sophisticated attacks. Therefore, innovative and robust authentication mechanisms are needed. Continuous authentication is an approach that goes beyond traditional single-point authentication, monitoring user behavior and constantly verifying access. Mouse movements include parameters about the user's mouse movements (speed, trajectory, type of operation, etc.). By analyzing these patterns, user authentication is performed.

And while related prior research has proposed a variety of approaches, including biometrics and behavior-based authentication, mouse dynamics has emerged as a non-intrusive yet effective means. This approach identifies and authenticates users based on their interaction patterns.

The study examines user behavior patterns in different gaming environments and their impact on the performance of machine learning models. It assesses the effectiveness and versatility of continuous authentication methods based on mouse dynamics. A comprehensive overview of previous studies is also provided, illustrating the different approaches and their results.

Proposed Method

During the data collection phase, 19 college students participated and their mouse movements were recorded in different games (Poly Bridge and Team Fortress 2). These games were chosen to demonstrate static strategies and action behaviors. Data were collected on standardized hardware and characteristic mouse movement patterns were observed.

The study design then executed a raw data to structured data processing pipeline, which included feature extraction and data normalization. Feature extraction included parameters such as mouse movement speed and click patterns, and feature sets were selected for model development. In addition, models such as GRU, LSTM, Decision Tree, and Random Forest were evaluated and the best performing model was finally selected.

Finally, model evaluations used AUC and ROC curves to assess model performance in light of data imbalances; F1 scores were also used to provide a complementary assessment of model accuracy and reproducibility. Sequence data were flattened in the case of Decision Tree and Random Forest models, and the study combined diverse methodologies to provide a comprehensive evaluation.

Experiment

Team Fortress 2: The GRU and LSTM models show high generalization ability, with test scores reflecting training scores. On the other hand, the DT and RF models show perfect scores on training data, but tend to perform poorly on test data. In particular, the RF model shows stable performance on test data, suggesting a low risk of over-fitting.

Poly Bridge: the GRU and LSTM models maintain stable high performance with slightly lower test scores. On the other hand, the DT and RF models show perfect scores on training data, but tend to perform poorly on test data. In particular, the DT model shows much lower performance on test data and is likely to be overfitting.

Team Fortress 2 and Poly Bridge integration: the GRU and LSTM models show high generalization ability and are effective for user authentication in both game environments; the DT and RF models show perfect scores on training data, but tend to perform poorly on test data. and the DT model in particular has a high risk of over-fitting.

consideration

The purpose of the study was to explore the possibility of using mouse movements in two different gaming environments as a means of continuous authentication. The results showed that the dynamics of individual mice are consistent and provide a reliable indicator of user authentication.

It is also important in that it provided a more comprehensive analysis than previous studies, covering both mild and intense sessions. It also highlights similarities with existing literature and competitive performance.

And the LSTM and GRU models showed competitive performance, while the DT and RF models also demonstrated robustness. The RF model in particular highlighted a balance between accuracy and repeatability, suggesting a high potential for practical implementation.

The results of the study highlight the versatility of the model used, with mouse dynamics providing a viable method for ongoing certification. Not only do they meet the high standards of previous studies, they also demonstrate the potential for practical implementation.

Conclusion

This study confirms the potential for continuous authentication using mouse movement dynamics. The dynamics of individual mice were shown to be consistent and useful for user authentication in different game scenarios. This study extended the scope of previous studies by covering quiet to active sessions. Results were consistent with and outperformed existing studies. The results of this study confirm that mouse movement dynamics is a useful tool for continuous user authentication, demonstrate the utility of the model used, and show the potential for future behavioral biometrics.