Keystroke Dynamics Based User Authentication using Numeric Keypad Baljit Singh Saini Research Scholar: CSE,Sri Guru Granth Sahib World University, Fatehgarh Sahib, Punjab Asst. Professor: CSE, Lovely Professional University Phagwara, Punjab Email: baljitsaini28@gmail.com Navdeep Kaur CSE,Sri Guru Granth Sahib World University, Fatehgarh Sahib, Punjab Email: drnavdeep.sggswu@gmail.com Kamaljit Singh Bhatia ECE, GZSCCET, Maharaja Ranjit Singh Punjab Technical University Bathinda, Punjab Email: kamalbhatia.er@gmail.com Abstract—Keystroke dynamics is the study to iden- tify/authenticate a person based on his/her typing rhythms, which are inferred from keystroke events like key-press and key-release. A lot of research work has been done in this field where the researchers have used either only alphabetic or alphanumeric or only numeric inputs. In this paper we address the question - What is the best possible numeric input for authentication using keystroke dynamics. We accomplished this by making the users enter four different numbers. Each number consisted of 8-digits. Out of these four numbers two were random numbers while the other two were formed using digits which had some pattern to them. Random Forest and Naive Bayes were used as classifiers. The results showed that using Random Forest classifier yielded best results when a random number is taken as input. The study also proved that a combination of hold time and latency as features yielded improved results. We achieved an average false acceptance rate of 2.7% and false rejection rate of 35.9%. I. I NTRODUCTION In modern times all the information is stored and shared using computers or mobile devices. With increased use of mobile devices the risk of theft of sensitive data has also increased. To protect data we use password but these passwords can be easily cracked by the hackers [2]. For better security, measures like finger scan, retina scan etc. are used which are a form of physical biometric. But these measures are very costly to implement. Keystroke dynamics is a behavioural biometric method which identifies the user on the basis of his/her typing pattern [1] . The characteristics of a keyboard has cognitive qualities [6] and hence can be used very effectively for identification purposes. This biometric system works by extracting features from the collected data. Then a classifier is used to build up the user profile. The same process is repeated while testing and if the profile matches the one in the database the user is authenticated otherwise not. The working of a biometric system is shown in the figure 1. The deployment of keystroke dynamics for authentication does not include any extra cost as you just require a keyboard for typing which is an integral part of computer system. There are two types of authentication in keystroke dynamics: fixed text and free text. In fixed text the input text is predefined and the user has to type the same text during enrolment and authentication time. In free text the user is free to type any text according to his liking during enrolment and authentication time, thus eliminating the need to remember passwords. Although free text seems a better choice yet accuracy rate with free text is low as compared to Feature Extraction Classification User Data Feature Extraction Classification Database (User Profile) Enrolment Authentication/Identification Build Match Yes/No Fig. 1: Biometric System [15] fixed text. To measure the efficiency of the system three common mea- sures are used: 1) False Acceptance Rate(FAR) - It is the count of how many times an imposter is accepted as an authorised user. 2) False Rejection Rate(FRR) - It is the count of how many times a genuine user is rejected as being an imposter. 3) Equal Error Rate(EER) - It is the value at which FRR is equal to FAR. In this paper we worked with fixed text input method. The input consisted of 8-digit numbers that were typed using the numeric keypad. Each user typed four different numbers. Two numbers were random numbers while two formed some pattern. The aim of the study is to determine whether a random number or a number having some pattern to it, acts as a better input for making user typing profile. The rest of the paper is organized as follows. In Section II we summarise the work done so far taking numbers as an input. Section III discusses about the problem and the approach that is being followed. Section IV details about the methodology adopted. We first discuss about the data collection technique, then about the features used and lastly about the classifiers that are used for analysis.In Section V the analysis and results are discussed in detail. The final section concluded the findings of the paper and touches upon the possible future work. 25 978-1-5090-3519-9/17/$31.00 c 2017 IEEE