Radio Frequency Identification (RFID) technology allows wireless interaction between tagged objects and readers to automatically identify large groups of items. This technology is widely accepted in a number of application domains, however, it suffers from data anomalies such as false-positive observations. Existing methods, such as manual tools, user specified rules and filtering algorithms, lack the automation and intelligence to effectively remove ambiguous false-positive readings. In this paper, we propose a methodology which incorporates a highly intelligent feature set definition utilised in conjunction with various state-of-the-art classifying techniques to correctly determine if a reading flagged as a potential false-positive anomaly should be discarded. Through experimental study we have shown that our approach cleans highly ambiguous false-positive observational data effectively. We have also discovered that the Non-Monotonic Reasoning classifier obtained the highest cleaning rate when handling false-positive RFID readings.