A novel hybrid binary whale optimization algorithm with chameleon hunting mechanism for wrapper feature selection in QSAR classification model:A drug-induced liver injury case study

High dimensionality is one of the main challenges in Quantitative Structure-Activity Relationship (QSAR) classification modeling, and feature selection as an effective dimensionality reduction method plays an important role in machine learning, particularly in fields such as chemometrics. In this paper, for feature selection in QSAR classification modeling, a hybrid whale optimization algorithm (WOA) with a chameleon hunting mechanism (HWOA-CHM) is proposed, and its binary version is used to find the best subset for wrapper feature selection in the QSAR classification model. First, a chaos weighting factor is introduced and used as a perturbation factor to increase the diversity of populations. Second, a retractable transformation strategy is designed to prevent the HWOA-CHM from falling into a local optimum. Third, the chameleon predation mechanism is introduced to improve the convergence accuracy of the HWOA-CHM. The performance of HWOA-CHM is evaluated and compared with state-of-the-art classical algorithms and well-known WOA variants. Then, a binary HWOA-CHM (BHWOA-CHM) was designed to solve the feature selection, the BHWOA-CHM is validated using the UCI machine learning repository and compared with binary version WOA, and well-known WOA variants in terms of accuracy, number of features, and time. Finally, BHWOA-CHM was used to solve the high-dimensional feature selection problem in the drug-induced liver injury classification model. It has shown excellent results in terms of feature selection compared to other methods. The proposed method effectively improves the robustness of QSAR predictions while reducing the complexity of the feature sets, demonstrating its potential for improving the accuracy of QSAR models.