Determination of performance to verify the synthetic identity theft by training the neural networks

摘要

This paper presents a method to analyse the various identities of a user and thus determine if any synthetic identity theft has been committed. Here three type of data is taken i.e., Input dataset (X), Normal dataset (Y) and Target Dataset (Z) are taken. The various identities used may be text or string data such as Candidate's Name, Date of Birth, Time of Birth Place of Birth, Home Address, Father's Name, Mother's Name, Husband's/ Wife's Name, Ration Card, Aadhar card Number, Voter's ID, Pan Card Number, SSLC Marks Card, Degree Proof, Blood Group, Face Image, Iris Image, Physical Features(Extra Thumb), Mole Marks, Injury Marks, Specimen Signature, Telephone Number, Mobile Number, Passport Number and Driving License Number. The various identities are classified in the category as 100% - High Identity with a correct information, 75% - Medium Identity with partial correct information and 30% - Low Identity with a wrong information. Each user are given the various score. The input values ranges from 0% to 100% for the various identity that is available. The normal values also ranges from 0% to 100%. The expected values are either 0% or 100%. With the above values training is given to the neural networks and the progress is obtained for the epoch values, time, performance, gradient and validation checks. The Performance, training state, confusion matrix and receiver operating characteristics are plotted for the plot interval of 9 epochs.