Rank-Based Adaptive Brooding in a Mimetic Coral Reefs Search for Feature Selection

Feature selection (FS) is an essential preprocessing step in machine learning and data mining. FS algorithms eliminate the irrelevant and redundant features from a feature vector, thereby increasing the accuracy and robustness of the data mining process while offering higher transparency and lower computation. Here, we suggest using the coral reefs optimization algorithms combined with a betahill climbing algorithm as part of a mimetic search strategy. The Coral Reefs is a population-based global search technique that combines well with local search techniques, as suggested here. However, such a combination is also computationally expensive. They could also converge prematurely with few iterations or insufficient population diversity. Accordingly, we propose here a more efficient search in the problem space by transforming the brooding operator into a function of the ranks of coral larvae. In this way, the chance of disturbing a larva with a higher rank is reduced, and at the same time, the exploratory role of larvae with a lower rank is used best. The performance of the proposed method has been evaluated using three KNN, Decision Tree, and SVM classifiers on eight UCI standard datasets using Precision, Recall and f1 score criteria. Results show that the rank-based adaptive brooding approach is superior to the fixed brooding approach in the high-dimensional problem space.