Comparison of SVM and Random Forest Methods for Online Signature Verification

Abstract

AbstractSignatures are widely used for authentication purposes. Signature verification has many applications in banking, in crossing international borders, in boarding of planes etc. For verifying identity of a person, signature is legally and widely accepted biometric trait. This work presents two simple and efficient methods for online signature verification. The paper proposes Support Vector Machine (SVM) and Random Forest Method for verifying online signatures. To measure the performance of algorithm f1 score is used and experiments were performed on SUSIG dataset. In Method-1, data after preprocessing is taken as feature set and in Method-2 feature vector are made by concatenation of bins of different attributes of the signature. The attributes taken are kth derivative of x and y coordinates, and kth derivatives of pressure. The classification is performed on extracted features using SVM and Random Forest. The performances of proposed methods were evaluated by using confusion matrix on SUSIG dataset. Results show that the proposed methods are capable of verifying online signatures with acceptable level of accuracy.KeywordsSVMRandom forestClassificationConfusion matrixFeature extractionOnline signature.