An Android Malicious Code Detection Method Based on Improved DCA Algorithm

Abstract

Abstract:Recently, Android malicious code has increased dramatically and the technology of reinforcement is increasingly powerful. Due to the development of code obfuscation and polymorphic deformation technology, the current Android malicious code static detection method whose feature selected is the semantic of application source code can not completely extract malware's code features. The Android malware static detection methods whose features used are only obtained from the AndroidManifest.xml file are easily affected by useless permissions. Therefore, there are some limitations in current Android malware static detection methods. The current Android malware dynamic detection algorithm is mostly required to customize the system or needs system root permissions. Based on the Dendritic Cell Algorithm (DCA), this paper proposes an Android malware algorithm that has a higher detection rate, does not need to modify the system, and reduces the impact of code obfuscation to a certain degree. This algorithm is applied to an Android malware detection method based on oriented Dalvik disassembly sequence and application interface (API) calling sequence. Through the designed experiments, the effectiveness of this method is verified for the detection of Android malware.

Research on optimization and application of Spark decision tree algorithm under cloud‐edge collaboration

“…From Figure 9, it can be concluded that the IDT algorithm proposed in this paper has significantly improved the accuracy of the algorithm compared with the traditional C4.5 algorithm. Compared with the algorithm in Wang et al, 10 the IDT algorithm proposed in this paper still improves in terms of accuracy. The smaller the redundancy between attributes in the data set, the more significant the accuracy improvement is.…”

“…Experiment 2: The improved C4.5 algorithm (Improved Decision Tree Algorithm for short: IDT algorithm) is validated with 10 times cross‐validation, and the average classification accuracy is used as a criterion for validation. The data set is the same as that of Experiment 1 and compared with the traditional decision tree algorithm and the SU_C4.5 algorithm proposed in Wang et al 10 The experimental results are shown in Figure 9.…”

“…Among them, the attribute subset S i contains k attributes, r ¯cf represents the mean value of the correlation between attributes and classes, and r ¯ff is the mean value of the correlation between attributes. From Equation (10), it can be derived that the meaning of its numerator is a measure of the ability to predict class attributes when the subset of attributes is all independent of each other, and the meaning of the denominator is a measure of the redundancy among the attributes. This equation is introduced into the C4.5 algorithm to transform the information gain rate so that the final selection of attributes takes into account the relationship between attributes and class attributes, as well as the relationship between attributes themselves, 46 so that a better attribute is selected as the division node of the decision tree.…”

“…In An and Shang, 8 Kendall's harmony coefficients were applied to the C4.5 algorithm, which allowed the algorithm to simplify the computation and solve the correlation between attributes, thus improving the accuracy of the C4.5 algorithm 9 . Wang et al 10 proposed a combination of the Relief algorithm and mutual information to find the appropriate set of conditional attributes to solve code odor in software structures 11 . (2) Improvements to decision tree algorithms in big data platforms include: Mu et al 12 utilized Pearson correlation coefficients to obtain the optimal partitioning attributes in constructing decision trees by a new feature metric, which was implemented in the Hadoop platform.…”