Shapley values has established itself as one of the most appropriate and\ntheoretically sound frameworks for explaining predictions from complex machine\nlearning models. The popularity of Shapley values in the explanation setting is\nprobably due to its unique theoretical properties. The main drawback with\nShapley values, however, is that its computational complexity grows\nexponentially in the number of input features, making it unfeasible in many\nreal world situations where there could be hundreds or thousands of features.\nFurthermore, with many (dependent) features, presenting/visualizing and\ninterpreting the computed Shapley values also becomes challenging. The present\npaper introduces groupShapley: a conceptually simple approach for dealing with\nthe aforementioned bottlenecks. The idea is to group the features, for example\nby type or dependence, and then compute and present Shapley values for these\ngroups instead of for all individual features. Reducing hundreds or thousands\nof features to half a dozen or so, makes precise computations practically\nfeasible and the presentation and knowledge extraction greatly simplified. We\nprove that under certain conditions, groupShapley is equivalent to summing the\nfeature-wise Shapley values within each feature group. Moreover, we provide a\nsimulation study exemplifying the differences when these conditions are not\nmet. We illustrate the usability of the approach in a real world car insurance\nexample, where groupShapley is used to provide simple and intuitive\nexplanations.\n