Faria-2017a

PhD thesis

This work sought to develop the cellular frustration model for computer security applications. In this sense, the required processes to materialize the cellular frustration model in a semi-supervised anomaly detection algorithm were developed. The discrimination capability of the cellular frustration algorithm was then compared with the discrimination capability of state of the art algorithms, namely support vector machines and random forests (SVMs and RFs, respectively). In the studied cases it is observed that the cellular frustration algorithm exhibits comparable, if not better, anomaly detection capabilities. Optimizations to reduce the high computational cost that rely on new computational paradigms, i.e. by the use of graphic cards, as well as optimizations to reduce the algorithm complexity were also described. In both cases it was observed a reduction of the computational time required by the algorithm. Finally, it was verified that the introduced improvements allowed the anomaly detection capability of the algorithm to become less sensitive to the perturbation of its parameters.