Abstract:
The increase in the use of mobile devices for various personal and sensitive purposes has
necessitated the development of several security measures to ensure safety of the devices, and most especially the data in them. With this background, the ability to identify
individuals and authenticate their credentials is key and the use of biometric technologies
is an important tool. In order to ensure security of data on mobile devices, keystroke
dynamics were looked into because of the low cost of implementation, computational speed and its non-invasiveness as it can be captured without necessarily informing the subject.
This research sought to address the issue of securing data on mobile devices by employing some Machine Learning algorithms. The K-Nearest Neighbours (KNN), Decision Trees (DT) and Multinomial Logistic Regression (MLR) classification algorithms were used to train and test the typing patterns of several individuals who volunteered to type a static passphrase. In the course of carrying out these experiments, the predictions given by Decision Trees were the most accurate of all the three base classifiers used. For instance after evaluation, DT had precision value of 99.92%, recall of 99.92%, F1 of 99.92% and 99.92% accuracy while KNN had precision value of 99.12%, recall of 99.12%, F1 of 99.12% and 99.12% accuracy; and MLR had precision value of 67.59%, recall of 67.54%, F1 of 67.55% and 67.54% accuracy. Afterwards, the Majority Voting Method was used as an ensemble to maximize the effectiveness of the three algorithms. The following values were recorded after evaluating the predictions from the Majority Vote Ensemble: precision value of 99.92%, recall of 99.92%, F1 of 99.92% and 99.92% accuracy. At the end of the research, evaluation revealed that the ensemble enhanced the individual performances of the base classifiers, thereby affecting the overall system which can be used for authenticating users of mobile devices
