Manufacturing Cost Prediction Through Data Mining

This paper explores the feasibility of predicting manufacturing costs of a product with nonparametric methods through data mining. A dataset consisting of qualitative and quantitative attributes was collected from one of the largest menswear manufacturers in North America. Products with similar characteristics in this dataset were analyzed and clustered together using the k-medoids algorithm and a cost estimation regression model was established for each cluster. This allowed an assessment of the best cluster to assign a new product and then predict its manufacturing cost based on its regression model. Rousseeuw’s silhouette width was applied to determine the optimal number of clusters as 18 and then polynomial regression models were assigned to determine the manufacturing cost of each cluster. The k-medoids clustering algorithm proved to be compatible with the dataset which included mixed categorical and numerical variables. Predictive costing through data analysis will lay the foundations for properly understanding and structuring manufacturing costs.