Altı uzay öğrenme tabanlı tek sınıf sınıflandırma kullanılarak optimize edilmiş bir kötü amaçlı yazılım tespit tekniği

Advancement in technology have imposed on security specialists to constantly find new methods and solution to system security problems. Accordingly, Cybersecurity was at the forefront of the solutions that must be researched. The problem here is that the danger exists continuously and has different forms that can be altered, this can camouflage the risk and thus being able to deceive the protection program. Since including the whole set of security problems and trying to find a solution that solves them collectively will be an impossible mission, since the security problems have a wide spectrum and covering them in one shot is impossible. So, taking each problem independently and covering it will be a better solution. Malware is considered as one of the oldest security threats that had widely affected systems, malware variations are numerous, covering the threat of malware will mean covering a good percentage of the security problems, since many of the problems depends on malware to implant its malicious applications inside the system. Three different strategies have been implemented; the goal was to detect malware. The start was with traditional Artificial Intelligence, aiming to test the influence of the header features that exists in the portable executable file on the accuracy of malware detection, two of the three strategies were implemented regarding this issue. The attained results have outperformed other similar implementations. The third strategy that has been implemented was a different novel strategy that has been implemented for the first time. The main aim of this strategy was to overcome all the minor wholes that may lead to obtain a faked good result and this was an important issue. Dataset imbalance and curse of dimensionality were an obstacle that may affect the attained results. This new strategy has bypassed these obstacles and outperformed any previously implemented strategy. The strategy was the subspace learning. The techniques implemented from this strategy were Subspace Support Data Description (SSVDD) and Graph Embedded SSVDD, before implementing them, One-Class Classification and Support Vector Data Description were also tested, results attained from subspace learning have outperformed other results from other techniques implemented previously.