A method and system for detecting tampering of an energy meter can involve, in a first phase: gathering magnetic field sample data, an average magnetic field strength, a minimum magnetic field strength, and a maximum magnetic field strength from a magnetic field condition applied to an energy meter by a magnetic sensor that measures a magnetic field in one or more directions, and using learning coefficients calculated from the magnetic field sample data and the average magnetic field strength, the minimum magnetic field strength, and the maximum magnetic field strength to classify with a classifier, magnet tampering conditions with respect to the energy meter. In a second phase, a magnet tamper event can be identified with respect to the energy meter when the magnet tampering condition classified by the classifier is greater than a magnetic detection threshold.