Internatilonal Symposium for Production Research, Viyana, Austria, 13 - 15 September 2017, pp.6
Data mining which has attracted great interest in recent years is the process of extracting meaningful, accurate, interesting and hidden information from the databases. Among the various data mining tasks classification is the most widely used one which aims to predict the classes of future data objects. Classification has been successfully applied to many areas and one of these areas is the production sector. By analyzing the production-related data, a lot of knowledge can be achieved and productivity of the production firms can be increased. Artificial neural networks are one of the most promising classification methods with high accuracy rates. An artificial neural network is a mathematical or computational model based on biological neural networks. The conventional algorithms used for training neural networks are based on mathematical computations, are slow and usually trapped into local optima. In order to overcome these disadvantages meta-heuristic algorithms are successfully used in the literature for training neural networks. In this study, a new generation meta-heuristic algorithm Stochastic Fractal Search is adapted for training feed-forward neural networks in order to determine the causes of quality defects in a fabric production facility in Kayseri. Stochastic Fractal Search algorithm is inspired by the natural phenomenon of growth and uses a mathematical concept called fractal. The quality defect dataset consists of 2461 instances, 9 categorical, 8 continuous inputs and 5 classes which express the quality groups. Artificial neural network is trained with Stochastic Fractal Search algorithm and accuracy of the algorithm is compared with some classification algorithms from Weka data mining software and TACO-miner algorithm proposed by Özbakır and her colleagues. Obtained results and comparisons have shown that the Stochastic Fractal Search algorithm gives high accuracy rates and capable of training artificial neural networks on quality defect dataset.