IOSR Journal of Computer Engineering (IOSR-JCE) e-ISSN: 2278-0661,p-ISSN: 2278-8727, Volume 22, Issue 3, Ser. II (May - June 2020), PP 13-19 www.iosrjournals.org DOI: 10.9790/0661-2203021319 www.iosrjournals.org 13 | Page A Comparative Study on Classifier Algorithms Based on Mouse Gesture Data for User Detection Masud Karim 1 , Md. Amir Hasnat 2 , Hasnain Heickal 3 , Md. Hasanuzzaman 4 1 Department of Computer Science & Engineering, University of Dhaka, Bangladesh 2 Department of Computer Science & Engineering, University of Dhaka, Bangladesh 3 Department of Computer Science & Engineering, University of Dhaka, Bangladesh 4 Department of Computer Science & Engineering, University of Dhaka, Bangladesh Abstract–In the age of information security, user detection methods are highly recommended for technological innovations. Continuous detecting a real user is a challenging issue. For this reason, researchers are giving importance on mouse gesture pattern to detect user. Different classifier algorithms are using in this technology. This paper triumphs a comparative study on classifier algorithms, Support Vector Machine, K-Nearest Neighbor and Naive Bayes based on mouse gesture for user detection process. Benchmark data of mouse gesture are collected and for our own testing more dataare captured by using Jitbit Macro Reader. Mouse gesture features are generated from both benchmark data and our collected data. Classifier algorithms are applied for these features to detect user mouse gesture pattern. It is found that K-Nearest Neighbor classifier shows the best performance to detect user. Keywords—Classifier Algorithm, Support Vector Machines (SVM), K-Nearest Neighbor (kNN), Mouse Gesture, Naïve Bayes, Behavioral Authentication. --------------------------------------------------------------------------------------------------------------------------------------- Date of Submission: 07-05-2020 Date of Acceptance: 21-05-2020 ------------------------------------------------------------------------------------------------------------------------------------- I. Introduction This paper presents a comparative data analysis of user detection from the experiences of mouse gesture data. In terms of highly secured smart system, information security management is dynamically changing. This paper will be a good reference for comparative analysis of different classifier algorithms on mouse gesture data for user detection. This study gives a primary introduction to mouse gesture features and examples of feature values. We have shown brief discussions on three classifier algorithms are Support Vector Machine, K Nearest Neighbor, Naïve Bayes. Then visualizes the resultant figures of all kinds of stimulated data analysis and describes the comparative study of different classifier algorithms in favor to this paper as well. Finally, the conclusion summarizes the results and gives the future projections for further research works. II. Related Work Chinmayee.KS et al. [1], have proposed a framework to authenticate user from mouse movement data that covers four modules: gesture creation, data acquisition and preprocessing, feature extraction and classification. They also worked on static authentication. They have analyzed mouse movement data by using Hidden Markov Model. The mouse movement data captured by asking user to draw a mouse gesture. First gestures are kept as template and next time user replaced gesture several times and these compared with template. The proposed system is recommended for complex environment like e commerce or e learning. Anam Khan et al. [3], have published a survey on performance analysis of mouse movement based user authentication researches. The resultant is compared based on FAR and FRR. Mouse dynamic features are shown during their study and also the basic features are identified. Algorithms and classifications are used in various related features those are briefly described. This paper shows how to using tools and techniques of using mouse movement based authentication. S. Suganya et al. [4], have addressed a method that creates a database of containing mouse dynamic data like co-ordinate value time stamp value and mouse operation as well. From these features, features vectors are achieved. The referred dataset contains static mouse behavior data of 20 users which is collected from available open source. By using diffusion map algorithm, they have proposed to reduce the dimension of future vectors. The author used neural network classes which prepared from the number of simple and highly interconnected processing elements. Performance analysis is shown by comparing existing proposed systems as well as FAR and