MPF-FS: A multi-population framework based on multi-objective optimization algorithms for feature selection
Jie Yang,Junjiang He,Wenshan Li,Tao Li,Xiaolong Lan,Yunpeng Wang
DOI: https://doi.org/10.1007/s10489-023-04696-0
IF: 5.3
2023-06-25
Applied Intelligence
Abstract:Feature selection algorithms based on evolutionary computation have continued to emerge, and most of them have achieved outstanding results. However, there are two drawbacks when facing high-dimensional datasets: firstly, it is difficult to reduce features effectively, and secondly, the "curse of dimensionality". To alleviate those problems, we take the initial population generation as an entry point and propose a variant initial population generator, which can improve diversity and initialize populations randomly throughout the solution space. However, during the experimental process, it was found that the improved diversity would cause the algorithm to converge too fast and thus lead to premature. Therefore, we introduced multi-population techniques to balance diversity and convergence speed, and finally formed the MPF-FS framework. To prove the effectiveness of this framework, two feature selection algorithms, multi-population multi-objective artificial bee colony algorithm and multi-population non-dominated sorting genetic algorithm II, are implemented based on this framework. Nine well-known public datasets were used in this study, and the results reveal that the two proposed multi-population methods on high-dimensional datasets can reduce more features without reducing (or even improving) classification accuracy, which outperforms the corresponding single-population algorithms. Further compared to the state-of-the-art methods, our method still shows promising results.
computer science, artificial intelligence