TY - INPR A1 - Schramm, Simon A1 - Pieper, Matthias A1 - Vogl, Stefanie T1 - Orthogonal Procrustes Based Anomaly Detection and Error Prediction for Vehicle Bills of Materials T2 - SSRN N2 - Industrial Bill of Materials (BOM) suffer from an surging complexity and cause errors in production which have detrimental effects on a product’s profitability. Currently, BOM anomalies have to be identified manually and errors have to be detected in the same way. This preprint describes a combination of data analysis and Machine Learning methods, such as hierarchical and agglomerative clustering, an isolation forest algorithm, association mining and a multi-output Artificial Neural Network, all based on a deterministic distance measure for an industrial BOMs. Solving the orthogonal Procrustes problem for complex, multi-level matrices, a distance measure for real world industrial BOMs was derived. A multi-output MLP was used in order to predict error probabilities with a time- reference. Our results show how to detect anomalies and predict errors in a complex, multi-level BOM based on historical, labelled data. While other authors focus on the mere comparison of BOMs, we aimed at a holistic approach, combining descriptive and predictive methods in order to forecast where in a BOM and at what time of BOM creation process errors occur. The resulting, prescriptive system was tested using real world data and has shown to effectively predict where and when BOM errors are probable to occur. Consequently, the prescriptive system is superior to prior, purely predictive systems, can help to decrease errors and thereby decreases product development time and cost in real world companies. KW - Orthogonal Procrustes KW - Bill of Materials KW - Isolation forest KW - Multi-output Multi Layer Perceptron KW - Association mining KW - Prescriptive modeling Y1 - 2022 U6 - https://doi.org/10.2139/ssrn.4120321 PB - Elsevier ER -