Affective Health is a mobile biofeedback monitoring system that measures galvanic skin response, pulse and movement, data which is sent through Bluetooth to the mobile phone where it is displayed on an interactive interface. The representation of the movement in the first versions of the system did not include any information about the type of activity the user performed. For an improved version of the system we have therefore tried to infer movement more precisely. A Naïve Bayes classifier was used for movement identification. The classifier was cross validated and tested on data obtained from 6 persons. We present quantitative results for different scenarios and selection of features and conclude that the proposed techniques indeed worked very well.