Hyper-parameter Optimization of Regularized-Extreme Learning Machine Using a Modified Opposition-Based Learning Artificial Bee Colony
Philip Vásquez-Iglesias, Amelia E. Pizarro, David Zabala‐Blanco, Juan Fuentes-Concha, Roberto Ahumada‐García, Paulo González · 2024
In this work, it is presented the Artificial Bee Colony (ABC) metaheuristic algorithm from a novel implementation of the strategy of Opposition-based Learning (OBL). The proposal is utilized in the optimization of hyper-parameters of a Regularized Extreme Learning Machine (R-ELM) by considering two popular datasets: Pima Indians Diabetes and Breast Cancer Winconsin. Hyper-parameters allow the network's operation to be adapted and, although it is possible to find the best combination with a simple brute force algorithm, the number of combinations can make an exhaustive process unfeasible in terms of time and/or energy cost, which makes the search for more efficient alternatives indisputable. Results show that the use of the proposed method allows the hyper-parameterization of the RELM in an average percentage of time less than 10% and an average performance greater than 99% compared to the brute force method.