Determining the factors that affect the production time in metal industry utilizing data mining methods

Abstract

Purpose The purpose of this article is to show that data mining methods can be used for identifying the factors that affect the production time in manufacturing industry. Then, this information can be used to take the measures for reducing the production time for especially companies that use the make-to-order policy. Theory and Methods: Rule-based and decision tree-based classification algorithms were used to build the data mining models for the identification of the factors. Based on data types, as rule based classification algorithms, part, decision table, jrip, oner; as decision-tree based classification algorithms, random tree, rep tree and lmt are chosen. Data preprocessing methods are applied to organize several production factors. To reduce the number of features several feature selection evaluators are applied to the data. This created several reduced data sets. For each data set and the full data set with all the features, models are built using the selected classification algorithms. Performance measures of the models are compered and the best model is selected accordingly. This methodology is applied to the production data of a machine manufacturing company which uses the make-to-order policy. Weka data mining program was used to build the models. Results: As a result of the study, it was determined that the factors that affect the production time the most are ambient temperature, product quantity, production month, part name, operator name, name of the machine produced, and cnc machine dimensions. The algorithm obtained with the best evaluation metrics result by using classification algorithms with before and after feature selection in the data set is random tree. The decision tree obtained with the random tree were evaluated and recommendations were given to the company as decision support input according to the rules determined to be crucial. Part, decision table, jrip, oner algorithms which are rule-based classification algorithms; random tree, rep tree and lmt were used among decision tree-based classification algorithms. Performance evaluation metrics as accuracy, roc curve, true positive, false positive, f-score obtained before and after feature selection were compared. Conclusion: The aim achieved by using 5 different feature selection evaluators and then 7 different classification algorithms, and the results obtained were compared. It is concluded that, the methods used are very successful in terms of classification evaluation metrics.