A novel approach to validate online signature using machine learning based on dynamic features

Abstract

This paper presents a new and mathematical method for online signature validation based on machine learning. In this way, the average values of the factors are taken into account to ensure validity. Here, seven different types of features used are x coordinates, y coordinates, time stamp, pen up and down, azimuth, height and pressure. Three new features are extracted from it, i.e., (displacement, velocity and acceleration) using the correlated extraction process to obtain dynamic feature of signature. These features are extracted from the popular dataset SVC2004. The extracted feature is then passed to various classifiers named as Naive Bayes, random forest, J48, MLP, logistic regression and PART. The result of genuine and forge signatures is obtained in terms of precision, true positive rate, false positive rate, F-score, etc. The obtained result is then compared with the existing method with respect to false acceptance rate and false rejection rate.