Abstrait

Authentication of User Based On Mouse-Behavior Data Using Classification

G. Muthumari, R. Shenbagaraj, M. Blessa Binolin Pepsi

To authenticate a user, classification based on mouse operating behavior is proposed. The data includes co-ordinate axes, time Stamp value and mouse operations. The holistic and procedural features are extracted from the mouse behavior data. The feature distance vector is calculated using Manhattan and Dynamic Time Warping method based on feature vector for representing the original mouse feature space. Kernel PCA method is used to reduce the dimensionality of the feature distance vector. Then the one-class Support Vector Machine classifier is applied on the distance-based eigenspace feature to analyze whether the input sample is legitimate user or an imposter. The performance of the proposed method is measured by False Acceptance Rate and False Rejection Rate. The test result proves that, the proposed method KPCA with one-class Support Vector Machine provides low error rates with good accuracy than the existing method PCA with classifier.