The dichotomy between users' privacy behaviours and their privacy attitudes is a widely observed phenomenon in online social media. Such a disparity can be mainly attributed to the users' lack of awareness about the default privacy setting in social networking websites, which is often open and permissive. This problem has led to a large number of publicly available accounts that may belong to privacy-concerned users. As an initial step toward addressing this issue, we examined whether profile attributes of Twitter users with varying privacy settings are configured differently. As a result of the analysis, a set of features is identified and used to predict user privacy settings. For our best classifier, we obtained an F-score of 0.71, which outperforms the baselines considerably. Hence, profile attributes proved valuable for our task and suggest the possibility of the automatic detection of public accounts intended to be private based on online social footprints.