A mobile malicious code classification method based on feature selection includes extracting Application Programming Interface (API) feature information including a package name, a class name, a method name and a description from a malicious application of a predefined category, vectorizing a training dataset generated using the package name, the class name and the method name in the API feature information for deep learning, learning the vectorized training dataset to generate a classifier, probabilistically classifying to fit a target malicious application into a category, and defining the category of the target malicious application using a result of the classification and outputting a classification important API. Accordingly, it is possible to deal with malicious behaviors of malicious applications quickly and prevent damage caused by the malicious behaviors.