We explore the applicability of machine learning techniques to insurance pricing. Because transparent, interpretable models are preferred, we start from single recursive trees, and we work towards more advanced ensembles such as bagged trees, random forests and boosted trees. Complete tariff plans are developed based on the frequency and severity of claims. The cost functions and performance measures used in the algorithms are adapted to the characteristics of claims data: count data with excess zeros on the frequency side and scarce, possibly heavy-tailed and right-censored data on the severity side. We also present visualization tools for assessing the importance of the different risk factors and their impact on the price of an insurance contract.