Intelligent detection method for interpretable malicious Android applications

The invention discloses an interpretable intelligent detection method for malicious Android applications. For a malicious sample, the classification model firstly gives a binary label of malicious detection, and then gives a specific explanation of the model; in the field of Android malicious detection, explanation is generally related to features; performing feature pruning by using a descending-order feature ranking matrix, generating an SVC model after feature engineering, using a full-zero vector as a reference baseline value of a DeepLIFT algorithm, and then calculating a preliminary score of each layer of features by comparing the difference between real input and the baseline value; and finally, generating an auxiliary matrix of the model by using RFE, recalculating the influence of the features on the model through the auxiliary matrix, and determining a final score. For each malicious sample, the method not only provides a classification result, but also provides reliable classification result interp