Enhanced Gold rush optimizer for Feature Selection Application in Software Fault Prediction Datasets

Abstract

The Gold Rush Optimizer (GRO) is a popular metaheuristic algorithm proposed recently. However, it has slow convergence, low accuracy, and easily gets stuck in local optima when solving real problems. To address these issues, we propose an enhanced version called the Enhanced Gold Rush Optimizer (EGRO). Our algorithm achieves better performance through four new mechanisms. First, a Sinusoidal Bridging Mechanism uses the sine function’s periodic waves to boost global search. Second, we create a new partner selection strategy based on Euclidean distance. Third, an adaptive Levy flight strategy dynamically adjusts the search step size and direction to improve population diversity. Fourth, a Metal Detector Strategy combines gradient feedback of the "M" factor to accurately avoid local optima traps. To test EGRO, we build a two-level evaluation system. Tests on the  CEC2022 benchmark sets show that EGRO performs best in convergence accuracy and stability for unconstrained optimization. In feature selection tasks using software defect prediction data, EGRO improves classification accuracy by 0.1\%–4.5\% compared to mainstream algorithms. Experiments prove that EGRO has strong scalability and practical value.