Optimization algorithms inspired by sea creatures in feature selection applications

In applications such as data mining, machine learning, and pattern recognition, feature selection is an essential preprocessing step. By removing redundant and irrelevant features, it increases its classification accuracy by removing redundant and irrelevant features while reducing the computational expense of the learning model,. In this thesis, nine nature inspired metaheuristic algorithms and one traditional method tested on twenty-two benchmark datasets taken from UCI database. Test results are evaluated using five different evaluation metrics including accuracy, precision, recall, F-Score and selection size. Four different classifiers are used to better understand the correlation between the algorithms and the classifiers. Algorithms are divided into three categories: algorithms inspired from sea creatures including Emperor Penguin Colony (EPO), Marine Predators Algorithm (MPA), Salp Swarm Algorithm (SSA), Jellyfish Search Optimizer (JS), Sailfish Optimizer (SFO) and Whale Optimizer (WOA); well-known algorithms in the literature including Ant Colony Optimizer (ACO), Genetic Algorithm (GA) and Particle Swarm Algorithm (PSA); traditional method as Chi-Square (CHI2). In this thesis, we propose JS algorithm as binary JS algorithm to be used for the feature selection process for the first time in the literature. Test results showed that EPO is the best algorithm using naïve bayes classifier, MPA is the best algorithm using support vector machine (SVM) classifier, SFO is the best algorithm using K-Nearest Neighbor (KNN) classifier and JS is the best algorithm using random forest (RF) classifier for the feature selection process.