In the context of intelligent maintenance, spare part prediction business scenario seeks promising return-on-investment (ROI) by radically diminishing the hidden costs at after-sales customer services. However, the classification of class-imbalanced data with mixed type features at this business scenario is not straightforward. This paper proposes a hybrid classification model that combines C4.5, Apriori algorithms and weighted k-Nearest Neighbor (kNN) adaptations to overcome potential shortcomings observed at the corresponding business scenario. While proposed approach is implemented within CRISP-DM reference model, the experimental results demonstrate that proposed approach doubles the human-level performance at spare part prediction. This highlights a 50% decrease at the average number of customer visits per fault incident and a significant cutting at the relevant sales and distribution costs. According to best runtime configuration analysis, a real-time spare part prediction model has been deployed at the client's SAP system.