An Online neural network based password prediction, generation, and storage scheme
Mbaka, Winnie Bahati
The gradual change from traditional workplaces to online platforms has been attributed to shifting user requirements, economic factors, and lifestyle differences. Perhaps the most significant factor attributed to this change may be the advent of the 2019 outbreak of the Coronavirus pandemic making the topic of physical interaction among some of the severely affected aspects of life. To remedy this situation, all knowledge and employment institutions adopted various online platforms as a means of maintaining a continued learning and working processes. However, these technical advances presented the issue of upholding information integrity of individuals accessing materials over the Internet as they were required to authenticate themselves prior to gaining access to secured resources. However, authentication processes such as the use of passwords are prone to guessing attacks, one of the biggest challenges in modern computing. Such attacks occur because of the vulnerabilities of human-chosen passwords. Research indicated that despite innovation on other safer authentication mechanisms, passwords continue to dominate the authentication space because they are memorable, free and user-generated. In view of the above shortcomings, this study sought to develop an online scheme that is geared towards helping Internet users, generate stronger passphrases based on how predictable their preferred passwords are. To understand the underlying technologies in the creation of stronger passwords, the study analysed existing literature on the character composition of human-created passwords and available tools that can be used to perform predictive analysis and generation of complex secret words. Additionally, password managers were studied to realise their functionality in securely storing complex passphrases. Analysis of the findings of the research asserted the need to incorporate neural networks, integrated data-driven insights, and derived concepts from the Markov chain model in the development of an online password predictive and generative scheme with an embedded password manager that allowed users to store the complex secret words. The resulting accuracy score after the scheme was trained using 50 epochs stood at 0.90332413, equivalent to 90.3%.
A Thesis Submitted in partial fulfilment of the requirements for the Degree of Masters of Science in Information Systems Security at Strathmore University
Passwords, Neural networks, Markov Chain Model, Predictive analysis