Secretary bird optimization algorithm incorporating independent thinking mechanism and sine-square step length for feature selection.

An improved Secretary Bird Optimization Algorithm (ISSBOA) is proposed. First, an independent thinking mechanism (IM) enhances the algorithm's ability to avoid local optima traps and broadens global exploration during the optimization process. Second, a sine-square step size mechanism (SM) dynamically adjusts the search step size, effectively balancing the performance deficiencies of the Secretary Bird Optimization Algorithm (SBOA) in both the exploration and exploitation phases. To validate the effectiveness of ISSBOA, simulations are conducted on the IEEE CEC2017 benchmark test suite, with comparisons made against 7 classic metaheuristic algorithms and seven recently proposed improved algorithms. The results demonstrate that ISSBOA achieves optimal performance in two sets of comparison experiments: when compared with the 7 standard algorithms, ISSBOA outperforms them in terms of average fitness value on 23 out of 30 test functions and in terms of variance on 16 functions, achieving an average Friedman test rank of 1.80 (securing first place); when compared with the 7 high-efficiency improved algorithms, it excels in average fitness value on 19 functions and in variance on 15 functions, with an average Friedman test rank of 2.73 (ranking first). This indicates that the proposed ISSBOA has both high optimization accuracy and strong robustness. Additionally, an adaptive transformation function is used to convert the continuous-domain ISSBOA into a binary version (BISSBOA) for discrete optimization tasks such as feature selection. To validate the performance of BISSBOA, a comprehensive evaluation is conducted using 20 public datasets with different dimensions, and comparisons are made against 7 high-performance feature selection algorithms. The results show that BISSBOA outperforms the other comparative algorithms across five evaluation metrics, thereby confirming its practicality and superiority in the field of feature selection.