IoT devices authentication using artificial neural network

Abstract

User authentication is one of the critical concerns of information security. Users tend to use strong textual passwords, but remembering complex passwords is hard as they often write it on a piece of paper or save it in their mobile phones. Textual passwords are slightly unprotected and are easily attackable. The attacks include dictionary, shoulder surfing, and brute force. Graphical passwords overcome the shortcomings of textual passwords and are designed to aid memorability and ease of use. This paper proposes a Process-based Pattern Authentication (PPA) system for Internet of Things (IoT) devices that does not require a server to maintain a static password of the login user. The server stores user's information, which they provide at the time of registration, i.e., the R-code and the symbol, but the P-code, i.e., the actual password, will change with every login attempt of users. In this scheme, users may draw a pattern on the basis of calculation from the P-code and R-code in the PPA pattern, and can authenticate themselves using their touch dynamic behaviors through Artificial Neural Network (ANN). The ANN is trained on touch behaviors of legitimate users reporting superior performance over the existing methods. For experimental purposes, PPA is implemented as a prototype on a computer system to carry out experiments for the evaluation in terms of memorability and usability. The experiments show that the system has an effect of 5.03% of the False Rejection Rate (FRR) and 4.36% of the False Acceptance Rate (FAR), respectively.