Cognitive discriminative feature selection using variance fractal dimension for the detection of cyber attacks
The dimensions of data have increased significantly with the abundance of the data that we share. This high-dimensional data results in redundant and irrelevant features that creates challenges to existing machine learning algorithms. Redundant and irrelevant features slow down the training and testing process as a result affecting the performance and run time of a learning algorithm. Conventional feature selection methods showcase great potential to select important features. However, machine enabled feature selection models are unable to solve complex analysis in absence of human cognitive aspect to it. This thesis addresses the challenge concerned with reducing the data dimensions for machine learning algorithms from cognitive aspect. The reduction is carried out by a variance fractal-based complexity analysis to select a reduced set of features of a cyber attack dataset for each attack type that are discriminative in nature. Furthermore, integrated artificial neural network were created with inputs that comprised of the reduced features selected through the complexity analysis. A performance comparison is also provided using our proposed methodology with resulting minimized dataset features of each attack types with non-minimized dataset. The comparative analysis shows that the resultant discriminative features derived from our proposed method not only consume less resource but also speed up the training and testing process while maintaining good detection rates.
Cognitive Intelligence, Machine Learning, Fractals, Variance Fractal Dimension, Complexity Analysis, Feature Selection, Cyber attacks, Cyber security, Artificial Neural Network, Dimensionality, Network Threats, Cognitive computing, Computational intelligence, Discriminative features