Work-In-Process Decision Support System with Predictive Modeling in the Food Manufacturing Industry
Open AccessThe food and beverage industry has been considered as one of the essential industries in the United States and plays a critical role in not only the economy but also national security. Food manufacturing companies must achieve and maintain operational efficiency and reduce direct manufacturing costs in order to stay competitive and continuously produce and offer affordable foods to the public. One of the key contributing factors, which also has been recognized as an industrywise challenge, is how to effectively manage work-in-process (WIP) inventories to reduce and sustain WIP inventory levels without harming regular productions. Historically, WIP inventories in the discrete manufacturing process have been researched widely with fruitful results. However, similar research is rarely found in the continuous food manufacturing process. This research aims to introduce a framework for a decision support system imbedded with predictive modeling to predict WIP status using factors collected on the shop floor. It is highly desirable to identify critical factors and predict WIP status accurately so preventative measures can be proposed and take place to promote material recovery and reduce direct manufacturing costs. The raw data, after being cleaned, was split into the training, validation, and test sets. Due to data imbalance, two sampling techniques including up-sampling and synthetic minority oversampling technique, also known as SMOTE, were applied as remedy measures to tackle the issues caused by imbalanced data. The data was then processed by one-hot encoding so the eight classifier candidates, including multivariate adaptive regression splines, random forest, adaptive boosting, adaptive bagging, gradient boost machine, support vector machine, K-nearest neighbors algorithm, and logistic regression, were applied. The one-hot encoding process creates a large sparse matrix, in which each column represents one level of the categorical factors in the original dataset. Feature selection methods, including least absolute shrinkage and selection operator (LASSO), simulated annealing and the Boruta algorithm using random forest as the underlying algorithm, were applied to the sparse matrix to remove non-informative predictors to increase training speed and prediction accuracy. The alternative thresholds were determined using receiver operating characteristic (ROC) curves of the validation set. Three final candidates were recommended based on the prediction performance of the test set and total computation time. Majority vote, as a simple prediction ensembling method, was used to combine the final candidates to offset shortcomings. The confusion matrix and relevant assessment statistics demonstrated that the ensemble method performed better than individual candidates and improved the overall performance. The WIP decision support system can be deployed on the cloud using cloud service providers, such as Amazon AWS and Microsoft Azure. It has been demonstrated that the WIP decision support system imbedded with predictive modeling can accurately predict WIP status during the spoilage process. The system plays a vital role in promoting interdepartmental communication, improving material coverage, and reducing direct manufacturing costs, all of which lead an organization to be more cost-effective and competitive in the food industry.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.