Grey Wolf Optimization Parameter Control for Feature Selection in Anomaly Detection

Abstract

The performance of different mechanisms utilised to perform anomaly detection depends heavily on the group of features used. Therefore, dealing with a multi-dimensional dataset that typically contains a large number of attributes has caused problems to classification accuracy. Not all attributes in the dataset can be used in the classification process since some features may lead to low performance of classifiers. Feature selection (FS) is a good mechanism that minimises the dimension of high-dimensional datasets. Modified binary grey wolf optimization (MBGWO) is a metaheuristic algorithm that has been successfully used for FS. However, the MBGWO algorithm has a drawback in selecting sub-optimal feature sets from an original set of features. This drawback is related to the linearly decreasing value of a parameter where there is no control between the exploration and exploitation processes. This study proposed an enhanced binary grey wolf optimization (EBGWO) algorithm for FS in anomaly detection by controlling the balancing parameter. The new method focused on obtaining a value for a parameter that controlled the trade-off between exploration and exploitation. Evaluation of the proposed method was on the NLS-KDD dataset with different attack classes and compared with other benchmark algorithms, such as binary bat algorithm, binary particle swarm optimization, and four variants of grey wolf optimiser for FS. The experimental results indicated that EBGWO was superior than other algorithms, where it obtained 19 features only out of a total of 41 features with 87.46% classification accuracy. The proposed algorithm can be applied to detect anomaly in network intrusion and outliers in data that are significant but difficult to find.

Hybrid Gradient Descent Grey Wolf Optimizer for Optimal Feature Selection

“…In 2021, a method was proposed [ 27 ] to tackle anomaly detection problems by utilizing an enhanced grey wolf optimizer in feature selection of the multidimensional dataset by controlling the balancing parameter for exploration and exploitation. The parameter “ a ” is the main focus with its value increased or decreased depending on the current iterations' performance as compared to the previous iteration.…”

Improved discrete plant propagation algorithm for solving the traveling salesman problem

“…Single-based algorithms, including variable neighborhood search [13], simulated annealing [14], and guided local search [15], aim to improve a single candidate solution. By contrast, population-based algorithms maintain and enhance candidate solutions, often using population features to conduct direct search; such algorithms include biogeography-based optimization [16], grey wolf optimizer [17], particle swarm optimization (PSO) [18], emperor penguin colony [19], genetic algorithm (GA) [20], ant colony optimization (ACO) [21], black hole (BH) algorithm [22], and dragonfly algorithm (DA) [23]. In recent years, studies on plants have demonstrated that plants display intelligent behavior.…”

Performance evaluation of the particle swarm optimization for clustering based on different parameter selection

“…"Unsupervised" means that there is no predefined information about the form of the dataset. Generally, there are two ways to conduct the division of data into groups [2]: the deterministic approach, and the optimization approach. The former approach includes partitioned clustering, hierarchical clustering, density-based clustering, grid-based clustering, and model-based clustering [3].…”