Proceedings of StudentFaculty Research Day CSIS Pace University

Proceedings of Student-Faculty Research Day, CSIS, Pace University, May 1 st, 2015 Numeric-Passcode Keystroke Biometric Studies on Smartphones By Michael J. Coakley, John V. Monaco, and Charles C. Tappert

Abstract • Pace University Classification System was used to evaluate biometric data extracted from mobile phones • Feature data included: 1. Mechanical keyboard features 2. Touchscreen features 3. Combined mechanical & touchscreen features • Best results were associated with Touch. Screen biometric features

Relevance of Study • Use of mobile devices continue to climb dramatically – More mobile phones than people on the planet [12]. – Improved technology and capacity equates to more and more sensitive data being stored and access through mobile devices – Most devices are either secured via a small 4 character PIN or not securitized at all • Government interest and support – DARPA – National Institute of Standards and Technology (NIST) – National Science Foundation (NSF)

Related Work • Keystroke – Bakelman Dissertation at Pace University – Maxion & Killourhy – Maiorana – Trojahn & Ortmeir • Touch. Screen – Zheng – Kambourakis – Feng – Alariki

Mobile Device Biometric System • Android Bio. Keyboard – Virtual keypad developed on Android platform and used as the default keyboard on Android mobile devices – Text entry data captured on mobile devices – Data stored in SQLite Database – Data transmitted from devices to centralized server • Mechanical Keyboard Features – Data associated Key Press and Key Release events • Touchscreen Features – Screen coordinates – Pressure associated with each keypress

Data Collection • Devices – 5 identical Android LG-D 820 Nexus 5 Mobile devices – Virtual keypad capturing keystrokes • Participants – City of White Plains employees – Pace University Students (NYC & PLV) • • Each entered 10 digit string (914 193 7761) 30 times 58, 882 data records, 52 distinct participants 190 Keystroke & Touch Screen Features Data collected in two sessions several weeks – 44% Male, 55% Female, Avg Age = 23, 86% Right Handed

Data Analysis • Three Feature Sets processed by Pace Biometric Classification System (PBCS) 1. 2. 3. Mechanical keyboard features Touchscreen features Combined mechanical & touchscreen features • Plan to compare PBCS against other classifiers – e. g. , SVM • Classifiers will also be compared on CMU data • Platform – Hardware: 16 gigs RAM, 8 Cores (2 threads/core), 100 gig drive – OS: Linux – Pace Classifier: Python

Two Classification Procedures Used on the Three Feature Sets • Repeated Random Subsampling (RSS) – Max between size of 10/Max within size of 10 – 30 iterations • Leave One Out Cross Validation (LOOCV) – Full Dataset (No Sampling)

Equal Error Rates (EER): RSS vs. LOOCV RSS LOOCV Mechanical Keyboard Features 23% 20% Touch. Screen Features 13. 8% 4. 9% Combined Mechanical & Touch. Screen Features 14. 9% 7. 1%

Receiver Operating Characteristic (ROC) Curves for RRS Feature Data 100 90 80 70 60 Diagonal 50 EER = 23% EER = 13. 8% EER = 14. 9% 40 Keystroke Touch. Screen Combined 30 20 10 0 0 20 40 60 80 100

Receiver Operating Characteristic (ROC) Curves for LOOCV Feature Data 100 80 60 Diagonal Keystroke Touch. Screen Combined 40 20 EER = 20% EER = 7. 1% EER = 4. 9% 0 0 20 40 60 80 100

RRS versus LOOCV Mechanical Keyboard Features 100 90 80 70 60 Diagonal 50 RRS LOOCV 40 30 EER = 20% 20 EER = 23% 10 0 0 20 40 60 80 100

RRS versus LOOCV Touchscreen Features 100 90 80 70 60 Diagonal 50 RRS LOOCV 40 30 20 EER = 13. 8% 10 EER = 4. 9% 0 0 10 20 30 40 50 60 70 80 90 100

RRS versus LOOCV Mechanical and Touchscreen Features 100 90 80 70 60 Diagonal 50 RRS LOOCV 40 30 20 EER = 14. 9% 10 EER = 7. 1% 0 0 10 20 30 40 50 60 70 80 90 100

Conclusion • Study indicated that the Pace Classifier can be extended to authenticate data associated with and extracted from mobile devices • EER of 4. 9% associated with Touch. Screen features indicates that the classifier works very well with these types of features • Future research will include what adjustments can be made to improve performance on the combined feature datasets

THANK YOU