Predictive maintenance is vital in smart manufacturing because it can help reduce downtime and costs and enhance productivity and product quality. Preventive maintenance for computer numerical control (CNC) machine tools is crucial for performance optimization. The effectiveness of preventive maintenance is affected by factors such as the usage environment, maintenance plans, and records. Regular inspection and maintenance are necessary to address problems related to tool wear and spindle condition, which affect machining quality. Accordingly, artificial intelligence algorithms, including deep learning models, have been extensively used for predictive maintenance. Conventional feature engineering methods and Internet of Things-based machine health monitoring systems are effective in this domain. Recursive feature elimination is commonly used for feature selection, but it is computationally intensive. In this study, we established a hybrid approach combining simulated annealing (SA) and deep neural network (DNN) models for diagnosing and predicting potential failures or problems in smart production machines; this approach was noted to exhibit excellent performance in solving complex problems. Integrating SA and DNNs can enhance preventive maintenance, optimize CNC machining processes, and improve productivity and product quality in smart manufacturing. The key advantage of the proposed hybrid approach is its ability to optimize feature selection while reducing computational costs. Therefore, the approach has potential for advancing preventive maintenance in smart manufacturing and provides valuable insights for developing efficient production systems.