An adaptive smartphone anomaly detection model based on data mining

Abstract

With the popularization of smartphones, they have become the main target of malicious applications. In recent years, malware has become a major threat to Android smartphones. Detection for malicious applications in smartphone has become a research hotspot. There are many studies that were mainly based on host mode. Although this kind of approach has the advantage of an effective collection of client features, it can interfere with the original environment for detection and only applies to certain versions of phones. In this paper, we combine network traffic analysis with data mining to identify malicious network behavior. We improve Apriori algorithm to extract network traffic features from network data, and overall exposes malware functionalities through operational behavioral triggers. Then, it adopts a density-based local outlier factor (LOF) clustering algorithm to form a detection model. ADMDM model can effectively detect an anomaly, and it has good results in unknown anomaly detection. The proposed model can be used for daily smartphone security checking and evaluation. Moreover, ADMDM enriches techniques for dynamic smartphone behavior analysis.