A principal component analysis and risk assessment framework based on Projection Pursuit in ad hoc networks

Mobile ad-hoc networks (MANET) has highly dynamic topology, open access of wireless channel and unpredictable behaviors; however, absence of effective security mechanism render MANET more vulnerable to positive attacks. Conventional assessments always require large sample data satisfy specific distribution and establish models through subjective recognition, thus lack common applicability, objectivity and reliability. To solve this problem and make accurate assessment, we propose RAPCA-PP model on basis of Projection Pursuit theory to realize both risk assessment and attributes analysis. Due to Projection Pursuit's theoretical merits, RAPCA-PP is thoroughly data-driven; it can be applied to conditions with small sample quantity, incomplete data and no-prior experience. Using RAGA for solution, RAPCA-PP shows well convergence. Compared with Grey Relations Projection, it demonstrates both better accuracy and higher discrimination. Moreover, our model can analysis attributes by importance and eliminate redundancy. Experiment shows that assessment with eliminated attributes can also correctly reflect each node's performance. RAPCA-PP proved to be suitable for real MANET working scenarios.