Support Vector Machine Based Concurrency Anomaly Classification for Mobile Applications

Abstract

Mobile applications became integral part of our daily life as they provide excellent portability and convenience. Growing number of features in mobile applications enables diverse user actions, sensors data and information to be inputted using various operations. Since mobile applications are event-driven programs, it is very difficult to detect concurrency anomaly with existing methodologies and tool supports due to the non-deterministic executions and difficulties in reproducing execution sequences for testing. The existing techniques of concurrent anomaly detection are very limited as they are often focused on detecting a specific concurrent anomaly without providing a generic approach. Furthermore, they produce large number of false positive in their detection results. Therefore, we propose a novel technique to dynamically classify concurrent anomalies for event-driven applications. According to control flow graph and vector clocks, we manually generate the training and test examples for classification using support vector machine with Gaussian kernel. The support vector machine will effectively predict whether the application under test is anomaly or not. We also have conducted experiments to determined accuracy of the proposed methodology. As result, the accuracy of proposed prediction methodology reached 80% with 20% of false positive rate. When compared with other related works, the proposed methodologies with machine learning technique is proven to be more efficient and effective.