To enhance the decision support actions, interpret, and explain large volumes of data, a security analyst will leverage the functionalities of a Visual Intrusion Detection System (V-IDS). These systems will offer helpful insights from complex logs, provide a good overview of the entire system and they are often used as an extra anomaly detection mechanism.
Leading companies in cyber-security reveal that the human expertise in the loop of the cyber-security operations, especially for monitoring, detection and mitigation of the cyber-threat patterns is a necessity.
Dimensionality reduction (DR) is a procedure used to transform high-dimensional data into a lower-dimensional representation. Usually, this procedure involves the minimum number of features required to adequately describe the key properties of the data. It is mainly used with different types of real-world data (such as time-series, images, medical records, unstructured text, etc.) and enables processes of visualization, classification, compression, and others.
In the cybersecurity domain, applying dimensionality reduction to a dataset will help the V-IDS visualizing the data. Offering a comprehensive visualization of data in higher dimensions is not an easy task, so reducing the space to 2 or 3 dimensions allows the data to be plot and reveal patterns more clearly.
DR assists in providing a simple 2-dimensions (2D) or 3-dimensions (3D) visualization of all the obtained parameters forming patterns that can be easily interpreted by the system operator without significant knowledge requirements on data analytics.
Within the framework of data analytics, the key incentive of dimensionality reduction has been to offer the visualization of multi-dimensional data, usually achieved through various supervised and unsupervised algorithms. Recent breakthroughs in deep neural network (DNN) technologies have opened new horizons and extended the state-of-the-art in dimensionality reduction by providing new incentives to revisit classical solutions through the newly available deep learning tools.