TABBA: A Novel Feature Selection Method Based on Binary Bat Algorithm and T-Test

Discovering disease-related genes from high-dimensional biomedical data is currently a hot issue. However, most biomedical data has a large number of irrelevant or redundant features, which makes the data difficult to utilize directly. For this problem, an improved binary bat algorithm based on T-test and variable step size adaptive algorithm (TABBA) is proposed. The T-test is employed to generate the initial population. Then a variable step size adaptive algorithm is introduced to accelerate convergence and avoid falling into the local optimum. The performance of TABBA is compared with original binary bat algorithm (BBA), variable step size adaptive binary bat algorithm (ABBA) and other optimization algorithms. The results in four public biomedical datasets confirm that TABBA is superior compared to benchmark methods in term of accuracy and the number of selected features.