Recent developments in supply chains and supply chain management (SCM) lead to increased complexity and vulnerability of supply chain operations. Therefore, it is needed to better anticipate and prepare for or prevent disruptions from occurring. Increasing interest for the application of machine learning in supply chain (risk) management has been observed, but applicational studies are lacking. Therefore, a generalised methodology consisting of six steps is proposed to introduce and apply machine learning to provide insights in supply chain operations and predictive analytics for supply chain (risk) management. The six steps are: Data collection and exploration, Performance and metric definition, Data preparation and feature engineering, Supplier grouping and feature selection, Data pre-processing and Algorithm comparison and evaluation. The methodology is applied to a case study from Philips Magnetic Resonance and Image Guided Therapy factory with the focus on predicting delivery performance of supplier deliveries by means of classification. Experiments show that the application of the methodology led to successful model development for individual suppliers resulting in binary and multiclass classification models obtaining Matthew’s Correlation Coefficient (MCC) scores up to 0.9 accompanied with 98% accuracy, 100% precision and 83% recall and MCC scores up to 0.75 accompanied with 88% accuracy, 85% macro-precision and 80% macro-recall, respectively. Additionally, feature selection showed the possibility to assist root cause analysis to improve (internal) operations or supplier relations by identifying important characteristics relevant for observed delayed deliveries. However, noticeable differences in performance for individual suppliers is observed, indicating the need for additional research focussing on increasing overall prediction performance (in the specific case study) and therewith direct applicability in operations.