A Dynamic Risk Assessment Framework Using Principle Component Analysis with Projection Pursuit in Ad Hoc Networks

Conventional risk assessments in Ad-Hoc Networks always require sample satisfy specific distribution with large quantity and establish models through subjective judgment, these methods lack general applicability, objectivity and credibility. Moreover, some models only focus on single time point evaluation and failed to thoroughly reveal dynamic behavioral character. To solve these problems and make correct assessment, we propose RAPCA-PP model on the basis of Projection Pursuit theory to conduct both risk assessment and attribute analysis over dynamic time sequence. RAPCA-PP is completely data-driven and can be applied in small sample, incomplete data and no-prior experience. Compared with Grey Relations Projection, it boasts both better accuracy and higher discrimination. Since RAPCA-PP makes evaluation along time axis, it can reflect MANET's node series behavioral features. Experiments demonstrate that assessment with eliminated attributes can also correctly reflect each node's performance and be utilized for IDS realization. RAPCA-PP proved to be suitable for real MANET working scenarios.