Name
Title | ||
|---|---|---|
Equilibrium optimizer with divided population based on distance and its application in feature selection problems |
Abstract | ||
|---|---|---|
Effective machine learning relies on feature selection (FS) to preprocess the data to search for the best feature subset among all feature combinations, which is a global optimization problem (GOPs). The equilibrium optimizer (EO) is a physical-based meta-heuristic that is promising in solving global optimization problems but still exhibits unbalanced exploration and exploitation; low convergence accuracy; and can easily fall into local optima. This paper proposes an equilibrium optimizer with divided population based on distance (DEO) to improve EO’s global optimization abilities and to use it to solve FS problems. First, we propose an adaptive population division strategy with beta distribution, which makes DEO divide populations by particles’ distance weights and dynamic adjustment of the number of particles in the population, facilitating the algorithm to achieve a balance between exploration and exploitation. Second, we add distance factor to the exploitation population update to control the step size, increasing convergence accuracy. Third, the exponential term in the EO algorithm is introduced into the exploration population update as a disturbance factor to enhance the ability to jump out of a local optimum. Finally, twenty benchmark functions are used to verify the effectiveness of the improved strategy. Numerical experiments based on CEC’17 and FS experiments based on 14 UCI datasets are conducted to validate DEO’s global optimization and feature selection capabilities, respectively. Experimental results demonstrate that the proposed DEO significantly outperforms seven state-of-the-art algorithms and five other modified algorithms in terms of the optimization-seeking accuracy and convergence speed. Additionally, DEO also obtains overall better fitness as well as shorter running time than the other seven algorithms in FS problems. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1016/j.knosys.2022.109842 | Knowledge-Based Systems |
Keywords | DocType | Volume |
EO,Distance factor,Feature selection,Optimization problems,Algorithm | Journal | 256 |
ISSN | Citations | PageRank |
0950-7051 | 0 | 0.34 |
References | Authors | |
0 | 4 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Li Yu | 1 | 1509 | 116.88 |
| Wendong Wang | 2 | 821 | 72.69 |
| Jingsen Liu | 3 | 6 | 3.10 |
| Huan Zhou | 4 | 0 | 0.34 |