This article first analyzes and compares the advantages and disadvantages of various traditional data mining methods and machine learning algorithms. Then, a new model for identifying Android malicious code was proposed: an Android malicious code detection model based on Java code analysis and deep learning algorithms. The model first extracts Java source code from Android software packages. Furthermore, based on the object-oriented features of the Java language, an inheritance and cross reference relationship graph between various classes is constructed, and the graph is mapped to a grayscale image through its adjacency matrix. Finally, a CNN model is used for classification training to identify malicious programs. By analyzing and comparing the advantages and disadvantages of various feature extraction and selection methods. Using data cleaning and principal component analysis to remove features unrelated to classification, thereby reducing data complexity and improving model prediction accuracy.