Abstract:
|
This research centers on finding the statistical moments, network measures, and statistical tests that are most sensitive to various node degradations for the Barabasi-Albert, Erdos-Renyi, and Watts-Strogratz network models. Thirty-five different graph structures were simulated for each of the random graph generation algorithms, and sensitivity analysis was undertaken on three different network measures: degree, betweenness, and closeness. In an effort to find the statistical moments that are the most sensitive to degradation within each network, four traditional moments: mean, variance, skewness, and kurtosis as well as three non-traditional moments: L-variance, L-skewness, and L-kurtosis were examined. Each of these moments were examined across 18 degradation settings to highlight which moments were able to detect node degradation the quickest. Closeness was the most sensitive network measure and the mean was the most sensitive single moment across all scenarios. The results showed L-moments and L-moment ratios were less sensitive than traditional moments and that non-parametric tests were quite sensitive.
|