Skip to Main Content
In pattern recognition, Bayes classifier is considered a powerful tool in the decision making process as its gives the lowest probability of committing a classification error. This has been highlighted in a number of previous works. The diversity in algorithm parameters, criteria used and number of users involved as well as the evaluation method makes the task of comparing their results and selecting an appropriate system a very thorny one. The purpose of this paper is four fold: to study the different approaches reported using a true password, to establish how normalization and using a single variance/mean impact on the results, using a maximum number of features does necessarily improve performance and finally demonstrate that performance optimization is feasible through careful selection of features and approach taken.