Keystroke dynamics is a behavioral biometric modality which uses typing patterns on a keyboard to recognize individuals. The way of typing the password slightly changes with time, because of various factors (including user's training). This modification in the way of typing results in a decrease of performance recognition over time. In this paper, we analyse the correlation between the comparison score between a query, and a reference and the number of times the user has typed the password. After having quantified this correlation, we analyse the possibility of using stacked classification to take this aspect in consideration during authentication. Then, we verify if it is possible to classify users on their way of evolving their typing pattern. Results show that even if comparison scores are correlated with the number of time the user has typed the password, the use of a stacked classifier does not improve the results much.