A method and corresponding apparatus for automatically detecting and preventing fraudulent calls on a communication network. At least one example embodiment may include collecting CDRs on the communication network for a given time period, aggregating the plurality of call features for each of the collected call detail records by destination number, and utilizing machine learning to generate a decision model for determining if a destination number and/or a call to that destination number may be fraudulent. According to another aspect of the example embodiment, the decision model may be implemented on the communication network to detect and prevent fraudulent calls.