Machine learning-based product quality classification of the enterprise producing aluminum flat coil

A fundamental research challenge in organizations is determining product quality by constructing prediction models considering the parameters that affect product quality. By solving this problem, shipment planning can be organized quickly based on product quality status, the line can be stopped when necessary, or when it is expected that the required product quality will not be achieved, such products can be planned in the following productions production plan. When such forecasts are not made, the product quality can be recognized much later in the production process through quality control operations. Re-planning and rework expenditures are incurred as a result of this condition. This paper proposes a data-mining-based methodological framework for predicting the quality of flat coils produced by an aluminum flat coil casting company. In the study, the quality of aluminum flat coil was classified considering the data such as OperatorSidePrint, RollEntranceWaterTemperature, UpRollDiameter, DownRollDiameter, UpSpreyVelocity, DownRollExitWaterTemperature, SteelPlateLineVelocity, SteelPlateOperatorSideBending, DownSpreyBulk that affect the product quality by the use of K-means, CLARANS, BIRCH, Ward's Hierarchical Agglomerative Clustering, Clustering, and Logistic Regression, KNN, Artificial Neural Network, CART (Decision Tree), Random Forest (RF), Gradient Boosting Machines, Feed-Forward Neural Network, Naive Bayes, XGBoost and RF, Gradient Boosting RF and AdaBoost-LSTM ensemble models, and Simple RNN classification techniques. Following the application of algorithms to 23 different semi-finished products, RNN provided the best results in 35% of the semi-finished products, and AdaBoost-LSTM provided the best results in 18% of the semi-finished products.