A number of landmines buried around the world are substantial, and the damage is also occurring every year. In Korea, there is more than one occurrence of damage caused by landmines per year. To detect and remove these mines, the human uses metal detectors directly or unmanned robots in areas where suspected of mined land. However, not only landmines but also various metals can be buried along the mine, so it is difficult to determine whether a detected metal is a landmine or normal metal. In this study, the algorithm to determine whether a detected metal is landmine or normal metal was proposed using support vector machine (SVM) which is most commonly used for classification of machine learning models. Also, the effectiveness of this algorithm was verified through real experiments. For this purpose, electric current measurements sensor was mounted on a commercial metal detector for print out a changed current to numerical data when metal was detected. The comparison target of the experiment to verify the algorithm is a generic beverage can and a 9V square battery. The noise from the data produced by the sensors filtered out using filters. Through this, the SVM was learned in advance through the each 40 training data of the beverage can and battery, and the algorithm determined it whether the detected metal is the beverage can or battery when the metal detector detected the metal. If the detected data of a mine apply to the SVM, the proposed algorithm can determine that the detected metal is landmine or not.