Thin-film transistor liquid crystal displays (TFT-LCDs) are critical components in electronic product assembly. However, defects inevitably occur during panel manufacturing, and manual inspection for every defective panel is impractical. Current random sampling methods lead to excessive inspections without a systematic basis, resulting in unnecessary labor and inspection costs. This study aims to optimize the panel inspection process by collecting key defect-related parameters and developing an intelligent recommendation system leveraging big data analytics, statistical techniques, optimization methods, and information technology. The research methodology involves (1) data collection from automated optical inspection (AOI) machines, (2) records from initial inspection personnel, and (3) behavioral data of inspectors. After preprocessing, daily production quantities are used to determine batch sizes, integrating acceptance quality levels (AQL) with industrial audit tables to establish sampling plans. A decision tree model is then developed and validated, followed by rule-based updates incorporating historical data for system implementation. Results indicate that grayscale average brightness and Y-coordinates are significant features. By integrating sampling plans with machine learning, the system provides senior inspectors with real-time sampling decisions, including acceptance and rejection thresholds, and assigns risk scores to batches. Implementation of the system reduced the defect misclassification rate from 0.54% to 0.47% and the sample size from 300 to 161 panels, demonstrating its effectiveness in improving inspection efficiency.