Authenticating Passwords by Typing Pattern Biometrics

Passwords are a common measure used in Authentication systems to make sure that the users are who they say they are. The complexity of these Passwords is relied on while ensuring security. However, the role of complexity is limited. Users are forced to write down complex passwords since easy ones are easily guessed. This study aimed at evaluating the uniqueness of typing patterns of password holders so as to strengthen the authentication process beyond matching the string of characters. Using our own dataset, this research experimentally showed that k Nearest Neighbor algorithm using Euclidean distance as the metric, produces sufficient results to distinguish samples and detect whether they are from the same authentic user or from an impostor based on a threshold that was computed. Results obtained indicated that typing patterns are distinct even on simple guessable passwords and that typing pattern biometrics strengthens the authentication process. This research extends work in typing pattern analysis using k Nearest Neighbor machine learning approach to auto detect the password pattern of the authentic and non-authentic users. It also provides an investigation and assessment to the effect of using different k values of the KNN algorithm. Further to this field is the methodology for calculating an optimal threshold value with higher accuracy levels that acted as a basis for rejection or acceptance of a typing sample. Additionally is an introduction of a new feature metric of a combined dataset which is a concatenation of both the dwell and latency timings. A comparison of performance for independent and a combined dataset of the feature metrics was also evaluated.