Investigating Machine Learning Techniques for Solving Product-line Optimization Problems

Product-line optimization using consumers’ preferences measured by conjoint analysis is an important issue to marketing researchers. Since it is a combinatorial NP-hard optimization problem, several meta-heuristics have been proposed to ensure at least near-optimal solutions. This work presents already used meta-heuristics in the context of product-line optimization like genetic algorithms, simulated annealing, particle-swarm optimization, and ant-colony optimization. Furthermore, other promising approaches like harmony search, multiverse optimizer and memetic algorithms are introduced to the topic. All of these algorithms are applied to a function for maximizing profits with a probabilistic choice rule. The performances of the meta-heuristics are measured in terms of best and average solution quality. To determine the most suitable metaheuristics for the underlying objective function, a Monte Carlo simulation for several different problem instances with simulated data is performed. Simulation results suggest the use of genetic algorithms, simulated annealing and memetic algorithms for product-line optimization.