Development of regression tree and neighborhood-based methods for prediction problems: An application in the die

Making realistic and fast decisions in production and service systems gives companies a competitive advantage. Developments in information technologies provide companies with easy access to large amounts of data. However, estimating numerical values is one of the major challenges faced by companies. In this study, a methodology based on data mining is proposed for the solution of prediction problems. Tree-based and neighborhood-based methods are used in the proposed methodology. Tree-based methods are Regression Tree, Bagging Regression Tree, and Boosting Regression Tree. Neighborhood-based methods, K-The Nearest Neighborhood and Bagging K-The Nearest Neighborhood. Weighted estimation functions that take into account the local outlier factors, distances and the nearest neighborhood order of the objects in the data sets were used while creating the prediction models. It was aimed to increase the accuracy of the prediction models by performing an outlier analysis study. The performance of the proposed approaches was tested on nine comparative evaluation datasets. In the comparisons, it was observed that the ensemble methods developed by using weighted estimation functions after data preprocessing with outlier analysis increased the accuracy. In addition, a case study was conducted to estimate the die production times in a company that manufactures sheet metal dies. The performances of the developed models were evaluated using the data of 85 dies produced by the company between 2015-2018. Statistical results showed that the accuracy of the prediction increased with the proposed approach.