Multiobjective Harris Hawks Optimization With Associative Learning and Chaotic Local Search for Feature Selection

Abstract

In the classification problem, datasets often have a large number of features, but not all features are useful for classification. A lot of irrelevant features may even reduce the performance. Feature selection is to remove irrelevant features by minimizing the number of the feature subset and minimizing the classification error rate.So it can be regarded as a multi-objective optimization problem. Beacuse of its simple structure and easy implementation, Harris Hawks Optimization algorithm(HHO) is widely employed in optimization problems.In this paper, the multi-objective HHO is applied to address the feature selection problem. In order to improve the search ability of the algorithm, associative learning, grey wolf optimization and chaotic local search are introduced into it. An external repository is used to save non-dominant solution set. The results of feature selection on the sixteen University of CaliforniaIrvine (UCI) datasets show that the proposed method can effectively remove redundant features and improve the classification performance of the algorithm.INDEX TERMS multi-objective harris hawks optimization; feature selection; associative learning;chaotic local search

Feature Selection Problem and Metaheuristics: A Systematic Literature Review about Its Formulation, Evaluation and Applications

“…Measures the overall accuracy of a model in terms of the proportion of misclassified instances in a dataset. This metric has been used in [46,49,59,62,[64][65][66][67][68]70,71,93,120,122,131,133,134,140,150,179] and mathematically is defined as follows:…”

“…The reference point usually is the anti-optimal point or "worst possible" point in the objective space. This metric has been used in [62][63][64][65][67][68][69][70]73,75,120,179] (d) Inverted generational distance (IGD): Computes the average Euclidean distance from true Pareto fronts to its closest solution in the population. This metric has been used in [64,65,68,73,75] and mathematically is defined as follows:…”

“…In this systematic review, six non-parametric statistical tests were detected. (i) Wilcoxon signed-rank test or Wilcoxon test [29,36,48,50,59,78,79,85,88,90,91,95,99,108,110,115,119,121,[124][125][126]130,131,[136][137][138]142,148,151,152,159,165,169,170], (ii) Friedman test [39,57,59,75,77,94,96,107,108,120,128,132,136,140,142,154,…”

“…• Chaotic Maps-Search-Functions: With a total of 25 instances across the years [34,40,42,53,81,90,93,103,106,114,117,120,121,126,129,134,143,147,156,159,[161][162][163][164]183], this technique has seen consistent use, with a noticeable peak in 2022. Its application suggests that researchers find value in its chaotic dynamics to enhance the exploration capabilities of metaheuristics.…”

“…Local Search: This technique has been the most frequently employed, with a total of 28 instances [36,40,46,50,69,77,93,99,103,107,[111][112][113]120,[126][127][128][129][130]134,143,145,147,151,153,161,174,183]. Particularly in 2022, there was a significant surge in its application, indicating its effectiveness in refining solutions and improving convergence rates.…”