<p>This study investigates the real industrial challenge of complex demand scheduling with multi-delivery in parallel manufacturing environments. It focuses on high-mix, low-volume product families and multi-delivery production, where traditional scheduling methods often fall short. To tackle this, an Intelligent Classifier-based Scheduling Optimization Method is introduced. This data-driven multi-objective optimization method uses machine learning to characterize both production and demand features, enabling efficient and flexible scheduling. The explicit consideration of the following factors distinguishes this work from previous literature: (1) integrated production capacity allocation and reallocation, (2) leveraging production features to ensure line balance, and (3) data-driven cascaded classifiers with flexible feature fusion. The proposed method excels in generating rapid, high-quality schedules that minimize tardiness and changeovers, outperforming benchmark methods. It ensures timely deliveries while maintaining balanced production lines, even in the presence of capacity constraints. Furthermore, (4) it adapts to demand fluctuations by adjusting production capacity and schedules accordingly avoiding a complete rescheduling, minimizing the impact on allocated capacity. Industrial applications show that the proposed method significantly reduces tardiness and changeovers, highlighting its superiority in real-world manufacturing scenarios. Its flexibility and adaptability in handling varied demands demonstrate its potential to improve performances in parallel manufacturing.</p>