A multi-objective evolutionary algorithm for solving the feature selection problem of high-dimensional sparse data and its application in the absorption, distribution, metabolism, excretion and toxicity (ADMET) classification

A novel multi-objective sparse evolutionary algorithm (MOSEA) is proposed to solve the feature selection (FS) problem in high-dimensional sparse data in the medical field, which aims to balance the accuracy of ADMET classification with the number of features used during the drug design phase. Based on the architecture of evolutionary algorithms, MOSEA enhances search efficiency and convergence speed through an initial scoring strategy. It binarizes the real-valued vectors of the decision space using an X-shaped transfer function, applies a dynamic scoring strategy for sparsification, merges offspring and parent populations and uses an angular function for environmental selection in fitness evaluation, thereby optimizing the FS process. Compared with six multi-objective optimization algorithms, MOSEA has significant advantages in terms of the performance of the FS problem for high-dimensional sparse datasets and hepatotoxicity datasets, indicating that it can effectively solve the multi-objective FS problem of ADMET hepatotoxicity.