Determining the faulty and refund products in manufacturing system: application on a textile Firm

Purpose- In this study, it is aimed to improve the production and quality control processes of a company operating in the textile industry. For this purpose predicting faulty and refund products by using simulation of oversampling and undersampling applications. Methodology– In this study, there are 250 different variables and 72959 lines of data on the production line. These data have been taken from the last 1-year data of the firm. In this study, simulation has been done. New machine learning methods have been used by simulating. The reason for the simulation is that it was easy to detect the refund and faulty conditions made in a large lot group production in previous studies. However, the aim is to investigate whether the accuracy of the prediction algorithms will yield consistent results in terms of the increase in the number of refund and faulty products when production is made in a larger structure. In the simulation method, "oversampling" and "undersampling" methods have been used. While making simulation prediction, in the literature, boosting algorithms, which are used as ensemble machine learning techniques, have been used. In this study, simulation has been done as follows, while the number of production lots increased, refund and faulty products were increased within the same application. The reason for doing this is to investigate whether the prediction status in normal machine learning algorithms can be captured in a larger data stack. This process is called oversampling. Then, the "undersampling" method was applied. According to the “undersampling” method, it is aimed to determine the refund and defect situations in a smaller lot by taking samples of refund and defective products with less frequency. At the end of the study, the results were interpreted by applying boosting algorithms. Findings- As a result of the study, it is concluded that "undersampling" and "oversampling" simulations predict better than usual machine learning methodology. Conclusion- In this study, it has been observed that the ensemble machine learning algorithms (adaboost, xgboost, gradient boosting algorithms), which are one of the ensemble machine learning methods that emerged in 2016, were applied to the production data for the first time and showed success in the prediction of faulty and refund products.