Efficient Implementation of Improved Dragonfly Algorithm in Multi-Target Feature Selection

Abstract

High-dimensional data often contain many features that do not help the model and make training slower. This study uses a multi-objective dragonfly method to choose smaller feature sets while keeping good accuracy. The method applies crowding distance and a changing inertia value to guide the search. Tests on 15 datasets show that the method removes about 58% of the features and increases accuracy by 3.1% over the basic dragonfly version. The Pareto fronts cover a wider range of choices and become stable in fewer generations. These results show that simple changes in the search steps can improve both feature use and accuracy. The method is useful when small models are needed, but future work should test more classifiers, larger datasets, and more objectives.