Feature Selection Menggunakan Algoritma Meta-Heuristik

Abstract

Machine learning requires data to make predictions. Data can have a large number of features. The large number of features can cause machine learning models to overfit, increase model complexity, and high computational costs. Feature selection is one method for optimizing machine learning models. Feature selection reduces the number of features used in the learning process. This research proposes a feature selection method using meta-heuristic algorithms. The machine learning model serves as the objective function for the meta-heuristic algorithm. The objective function is evaluated at each iteration to obtain the most influential features in the model. The machine learning models used are Random Forest, k-Nearest Neighbors, and Support Vector Machine. The meta-heuristic algorithms used are Differential Evolution, Flower Pollination, Grey Wolf, and Whale Optimization. The research shows that using meta-heuristic algorithms can improve the accuracy of machine learning models with fewer features. The Support Vector Machine – Differential Evolution scheme has the highest accuracy and uses the fewest features.