TY - GEN
T1 - Improved Feature Engineering for Free-Text Keystroke Dynamics
AU - Abadi, Eden
AU - Hazan, Itay
N1 - Publisher Copyright:
© 2020, Springer Nature Switzerland AG.
PY - 2020/1/1
Y1 - 2020/1/1
N2 - Free-text keystroke dynamics is a method of verifying users’ identity based on their unique pattern of typing a spontaneous text on a keyboard. When applied in remote systems, it can add an additional layer of security that can detect compromised accounts. Therefore, service providers can be more certain that remote systems accounts would not be compromised by malicious attackers. Free-text keystroke dynamics usually involve the extraction of n-graphs, which represent the latency between n consecutive events. These n-graphs are then integrated with one of the various existing machine learning algorithms. To the best of our knowledge, n-graphs are the most widely used feature engineering for free text keystroke dynamics. We present extended-n-graphs, an improved version of the commonly used n-graphs, based on several extended metrics that outperform the traditionally used basic n-graphs. Our technique was evaluated on top of the gradient boosting algorithm, best performing algorithm on basic n-graphs and several additional algorithms such as random forest, K-NN, SVM and MLP. Our empirical results show encouraging 4% improvement in the Area Under the Curve (AUC) when evaluated on a publicly used benchmark.
AB - Free-text keystroke dynamics is a method of verifying users’ identity based on their unique pattern of typing a spontaneous text on a keyboard. When applied in remote systems, it can add an additional layer of security that can detect compromised accounts. Therefore, service providers can be more certain that remote systems accounts would not be compromised by malicious attackers. Free-text keystroke dynamics usually involve the extraction of n-graphs, which represent the latency between n consecutive events. These n-graphs are then integrated with one of the various existing machine learning algorithms. To the best of our knowledge, n-graphs are the most widely used feature engineering for free text keystroke dynamics. We present extended-n-graphs, an improved version of the commonly used n-graphs, based on several extended metrics that outperform the traditionally used basic n-graphs. Our technique was evaluated on top of the gradient boosting algorithm, best performing algorithm on basic n-graphs and several additional algorithms such as random forest, K-NN, SVM and MLP. Our empirical results show encouraging 4% improvement in the Area Under the Curve (AUC) when evaluated on a publicly used benchmark.
KW - Feature engineering
KW - Free text
KW - Keystroke dynamics
UR - http://www.scopus.com/inward/record.url?scp=85092191278&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-59817-4_6
DO - 10.1007/978-3-030-59817-4_6
M3 - Conference contribution
AN - SCOPUS:85092191278
SN - 9783030598167
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 93
EP - 105
BT - Security and Trust Management - 16th International Workshop, STM 2020, Proceedings
A2 - Markantonakis, Kostantinos
A2 - Petrocchi, Marinella
PB - Springer Science and Business Media Deutschland GmbH
T2 - 16th International Workshop on Security and Trust Management, STM 2020, held in conjunction with the 25th European Symposium on Research in Computer Security, ESORICS 2020
Y2 - 17 September 2020 through 18 September 2020
ER -