A key goal of after-sale services is to achieve customer satisfaction by providing a high-quality of service post-sale. In this context, repairable spare parts play a considerable role in balancing between service level and inventory value considering their relative cost compared to new buy parts. This article deals with forecasting the load of a repair centre in a closed-loop service parts supply chain. Since this supply chain is subject to variabilities and experts are modifying replenishment decisions, it is difficult to forecast this load and especially its peaks, this lack of visibility is creating difficulties to organise the repair centre. Thus, a methodology based on a hybrid machine learning method is proposed and then tested on real data from General Electric Healthcare. These tests showed a real improvement of the accuracy compared to simple machine learning methods or traditional forecasting methods.