One-handed Keystroke Biometric Identification Competition John V. Monaco 1 , Gonzalo Perez 1 , Charles C. Tappert 1 , Patrick Bours 2 , Soumik Mondal 2 , Sudalai Rajkumar 3 , Aythami Morales 4 , Julian Fierrez 4 and Javier Ortega-Garcia 4 1 Pace University, Pleasantville, New York, USA, {jmonaco,gperez,ctappert}@pace.edu 2 Gjøvik University College, Gjøvik, Norway, {patrick.bours,soumik.mondal}@hig.no 3 Tiger Analytics, Chennai, India, sudalai@tigeranalytics.com 4 Universidad Aut´ onoma de Madrid, Madrid, Spain, {aythami.morales,javier.ortega,julian.fierrez}@uam.es Abstract This work presents the results of the One-handed Keystroke Biometric Identification Competition (OhKBIC), an official competition of the 8th IAPR International Con- ference on Biometrics (ICB). A unique keystroke biomet- ric dataset was collected that includes freely-typed long- text samples from 64 subjects. Samples were collected to simulate normal typing behavior and the severe handicap of only being able to type with one hand. Competition participants designed classification models trained on the normally-typed samples in an attempt to classify an un- labeled dataset that consists of normally-typed and one- handed samples. Participants competed against each other to obtain the highest classification accuracies and submit- ted classification results through an online system similar to Kaggle. The classification results and top performing strategies are described. 1. Introduction Keystroke biometric applications have been investigated over the past several decades, attracting both academics and practitioners. There are commercial products available that analyze a sequence of keystrokes for human identification, or provide additional security through password hardening and continuous authentication REF. It is common to see er- ror rates below 10% for short text authentication [11], and below 1% in long text applications [12]. In terms of con- tinuous authentication, an intruder can accurately be iden- tified in less than 100 keystrokes [4]. While many perfor- mance evaluations are derived from normal typing behav- ior obtained in laboratory or natural settings, there has not been much research to determine how the performance of a keystroke biometric system degrades as a result of a user impairments, such as typing with one hand after having en- rolled with a normal both-hands typing sample. Such a sce- nario might be encountered in production or during a field experiments that impose little or no condition on how the system should be used. There are many performance-degrading scenarios that may be encountered during deployment of a keystroke bio- metric system. Variations in typing behavior can occur as a result of distractions, cognitive load, and sickness, to name a few. Consider the scenario in which a user has enrolled with normal two-hand typing and later restricted to typing with only one hand as a result of an injury or multitask- ing (e.g. using a desktop mouse with one hand while typing with the other). A robust keystroke biometric system should be able to handle this situation appropriately, although the correct response of such a system is not known at this point. Should the user be re-enrolled with a one-hand sample or can the user still be identified under this constraint? The results of this competition can help answer these questions. 2. Benchmark dataset A unique keystroke biometric dataset was collected from three online exams administered to undergraduate stu- dents in an introductory computer science course during a semester. Each exam contained five essay questions that required typing a response directly into a web page. Stu- dents took the three exams through the Moodle learning platform and their keystrokes were logged by a Javascript event-logging framework [1] and transmitted to a server. For the first exam students were instructed to type normally with both hands, for the second exam with their left hand only, and for the third exam with their right hand only. The benchmark dataset consists of 64 students who pro- vided at least 500 keystrokes on each exam. Approximately 1/3 of all exam attempts occurred in an electronic classroom on standard desktop computers to ensure instructions were followed when typing with just one hand. The remaining 978-1-4799-7824-3/15/$31.00 ©2015 IEEE ICB 2015 58