Research And Implementation Of Android Malware Detection Model Based On Deep Learning

At present,while Android phones bring convenience and quick services to people's lives,their user privacy leakage incidents also occur frequently.Android software,as a carrier of malicious behaviors,is urgently required for its security detection.The traditional machine learning model faces the detection data composed of multiple types of features.Due to the high data dimension and noise,Gu cannot effectively extract key information,and the deep learning method has a strong ability to learn the feature law It is suitable for processing such high-dimensional and complex multi-feature data.Based on this,this paper proposes an Android malware detection model based on deep shrinking noise reduction self-encoding network and a deep dual-pass convolutional neural network.In the research of Android malware detection model based on deep contractive denoising autoencoder network,first,reverse analysis of seven types of information such as APK file access permissions and sensitive APIs as feature attributes;then,using feature attributes as deep contractive denoising autoencoder network Input,use the greedy algorithm to extract the original feature information layer by layer from bottom to top to obtain the optimal low-dimensional representation;finally,use the back propagation algorithm to train and classify the obtained low-dimensional representation.Adding noise to the input data makes the reconstructed data more robust.At the same time,the Jacobian matrix is added as a penalty term,which enhances the antidisturbance ability of the deep autoencoder network.The experimental results verify the feasibility and efficiency of this model,and it has a high detection rate.In the research of Android malware detection model based on deep dual-pass convolutional neural network,first,digital data is visualized as image data;then,in order to extract more effective information,a convolution with two different sizes and working independently is designed.The kernel's deep dual-pass convolutional neural network model compresses the image through two dual-pass convolutional layers and pooling layers to obtain key information;finally,the Sigmoid classifier is used to classify the obtained low-dimensional information.The introduction of the Relu function avoids the disappearance of the gradient and speeds up the convergence speed.At the same time,the BN layer and Dropout method are added to enhance the generalization ability of the model.Experimental results show that this model can better obtain key information of feature data,the accuracy rate reaches 98.6% and the false alarm rate is only 1.19%.