On the effectiveness of risk prediction based on users browsing behavior