Análisis y detección de ataques informáticos mediante sistemas inteligentes de reducción dimensional

Abstract

This research work addresses the study and development of a methodology for the detection of computer attacks using intelligent systems and techniques for dimensional reduction in the eld of cybersecurity. This proposal is intended to divide the problem into two phases. The rst consists of a dimensional reduction of the original input space, projecting the data onto a lower-dimensional output space using linear or non-linear transformations that allow a better visualization of the internal structure of the dataset. In the second phase, the experience of an human expert is presented, which makes it possible to contribute his knowledge by labeling the samples based on the projections obtained and his experience on the problem. This innovative proposal makes a simple tool available to the end user and provides intuitive and easily interpretable results, allowing to face new threats to which the user has not been exposed, obtaining highly satisfactory results in all real cases in which has been applied. The developed system has been validated on three di erent real case studies, in which progress has been made in terms of knowledge with a clear guiding thread of positive progress of the proposal. In the rst case, an analysis of a well-known Android malware dataset is carried out, in which a characterization of the various families of malware is developed using classical dimensional reduction techniques. For the second of the proposals, it has been worked on the same data set, but in this case more advanced and incipient techniques of dimensional reduction and visualization are applied, achieving a signi cant improvement in the results. The last work takes advantage of the knowledge of the two previous works, which is applied to the detection of intrusion in computer systems on network dataset, in which attacks of di erent kinds occur during normal network operation processes.