Improvement of Committee Machine Performance to Solve Multiple Response Optimization Problems

Three phases are considered for multiple response optimization (MRO) problems. They are design of experiments, modeling and optimization. Committee machines (CM) as a set of some experts such as some artificial neural networks (ANNs) can be applied for modeling phase. Then, genetic algorithm (GA) determines the final solution with object maximizing the global desirability as optimization phase. That algorithm was implemented on five MRO case studies include target, minimizing and maximizing objects. Current article is a development of recent authors' work on application of CM in MRO problem solving. Initial approach in that work, includes a committee machine with four different ANNs. The CM weights are specified with GA which its fitness function was minimizing the overall RMSE for each response. In current work, a new approach was applies in finding the committee machine weights. The fitness function in this approach is made by minimizing the absolute error between CM responses and real data for each response, separately. A performance index is defined to evaluate different models performance. The results from five case studies show that there are noticeable decreasing in overall RMSE whereas there is a negligible decreasing in GD for new CM with respect to initial CM. this is due that less error is a confirmation of performance increasing for new committee machine. Index Term-Global desirability, Committee Machine, Multiple responses optimization, Genetic Algorithm