Android Applications Categorization
android projects titles topics ideas
With the development of Mobile Internet, the global mobile devices market has exploded in the past several years. Meanwhile, Android is the most popular mobile platform, holding nearly 82.8% of the global smartphone market share [1]. Mobile Internet brings convenience to our life. Nevertheless, it brings a lot of negative impacts to our life. The number of malicious software on mobile devices is growing and Android is the most affected mobile platform. According to 2015 MOBILE THREAT REPORT, 97% of malware application target to Android. In order to mitigate the threats on Android mobile devices, machine learning techniques have been widely used. https://codeshoppy.com/android-projects-titles-ieee.html According to whether to run the application during the detection process, the detection methods are divided into two kinds: static detection [2], [3], [4] and dynamic detection [5], [6]. Experiments show that when applying machine leaning technology to detect malicious software, adding the application category feature can significantly improve the performance and accuracy of the detection [7].
Thus, application category is important for android malware detection. However, when application publishers publish application to the Android application market, they need to manually select the category and they are more inclined to choose the category in the front row where you can see (e.g., if tools is in font of weather on the website, the publisher may choose tools even though the application is belong to weather). Besides, it’s difficult for application publishers to distinguish the difference between different categories for choosing the most appropriate category. More importantly, the malware application publisher may randomly choose application category to evade the detection of the application market. android projects titles topics ideas Therefore, an approach that can automatically categorize different types of applications can be helpful for both organizing the Android Market and detecting malicious applications. In [9] Shabtai Asaf et al. firstly used the data mining method to categorize Android applications. They extracted over 2200 features from Java byte-code and XML-files of Android apk, and they selected features using three selection methods: Information Gain, Fisher Score and Chi-Square. They obtained 89% of accuracy for classifying applications into only 2 categories: tools or games. Their limitation is that they just simply collected all the static features from the application but failed to conduct in-depth analysis to those features. Besides, the dataset they used is very limited — only contains 2 categories applications. In [10] Borja Sanz et al. proposed a new method to classify Android applications into several categories. They extracted features both from Android Market and the application itself. Their main contribution is to apply the data collected from Android Market into Android classification. However, they use the permissions declared in the manifest.xml rather than the permissions actual used in the application. Most applications have the phenomenon of over-claim of permissions [12] – the permissions declared in the manifest.xml file are not actually used in code. Therefore, the permissions declared in manifest.xml can’t completely reflect the real characteristics of the application. Given this background, we present a new methodbased on Bayesian classification, which utilizes several features extracted from Android Market and the application itself for classifying Android applications into several categories. The features we used include permissions, strings and description of the application in Android Market. Different from earlier works on Android categorization that rely on the permissions declared in Android manifest.xml [10], our approach uses the actual used permissions which are extracted by static analyzing APIs called by the application. Compared with previous work [9], [10], we take the category of an application into consideration which is determined by the app’s functions. Thus, we made an in-depth analysis of the string contained in application, and then extract the strings that can reflect its function as the string features, removing irrelevant string features, improving the accuracy of the classification. In summary, the main contributions of this paper different from existing related works in the literature are as follows: •Through analyzing the APIs called by the application, we extracted the actually used permissions, avoiding the error caused by the traditional use of the declared permissions. •Through analyzing and observing the usage characteristics of the strings in the applications, we extracted the string which can reflect the application functions as the strings feature target-oriented. •To build the model, we collected 13005 applications, which have been classified into 18 categories. As far as we know, compared with previous Android categorization experiments, ours has the greatest number and the most comprehensive categories of the datasets.