A Machine-Learning-based Approach for Parameter Control in Bee Colony Optimization for Traveling Salesman Problem

Metaheuristics are a set of algorithms which is capable of solving Combinatorial Optimization Problems (COPs). When a metaheuristic is used to solve a COP, one of the major aspects is to determine an appropriate parameter setting. Poor practice of determining parameter values may lead to inability to find optimal solutions or getting invalid conclusions from the experimental results. This research proposes a machine-learning-based parameter control mechanism for a metaheuristic, i.e. the Bee Colony Optimization (BCO) algorithm. The proposed mechanism consists of three main phases: Data Collection, Model Training, and Deployment. In order to examine the performance of the BCO algorithm with the parameter control mechanism, a set of 16 TSP instances is used as test bed. The experimental results show that it is significantly better than the BCO implementation using the parameter values that are determined via a manual tuning process. The proposed parameter control mechanism overcomes the shortcomings of manual parameter tuning and dynamically adjust the parameter values throughout the BCO optimization process.