The purpose of this study is to evaluate a hybrid data-driven framework for the simultaneous optimization of equipment failure risk cost and planned maintenance costs, relying on the output of the superior machine learning-based failure prediction model. The research is applied and data-driven, and the data used in this study includes 400 weekly records extracted from an 8-year period related to equipment maintenance activities in a three-tier coke production supply chain, from coal mining to the processing stage. In the first stage, equipment failure probability prediction was performed using several machine learning algorithms. Then, the output of this model was utilized as the input of a multi-objective genetic algorithm to extract optimal scenarios for preventive maintenance scheduling. The findings showed that the gradient boosting model has a higher capability in identifying complex equipment failure patterns, and linking its output to the genetic algorithm enables the extraction of a set of trade-off optimal solutions between reducing failure risk and controlling maintenance costs. The resulting Pareto front indicates that increasing maintenance intensity does not necessarily lead to minimizing the total cost, and selecting the optimal scenario depends on managerial priorities and the organization’s risk tolerance. The present study demonstrates that integrating machine learning-based failure prediction with multi-objective optimization can upgrade maintenance decision-making from a reactive level to a proactive and intelligent level.