光學組件的貼合是決定電子產品性能與結構品質的關鍵前製程，該製程涉及人機協作以滿足客戶的各種需求。在此類的生產製程中，若僅追蹤單一零組件的品質，將無法有效發現異常。因此，本研究以工單作為分析層級，以進行早期異常偵測。從平面膜層貼合與軌跡式塗膠製程中收集的兩年度資料顯示，這兩個製程的不平衡率分別為32.97與45.53，故為典型的不平衡資料。為了克服類別分佈懸殊導致的分類偏見，本研究首先採用過抽樣方法SMOTE-NC來平衡資料集中的類別分佈，接著為決策樹學習（Decision Tree Learning）與隨機森林（Random Forest）設定誤分類成本，以改善對異常工單的分類預測表現。
實驗結果顯示，以 F-measure 作為評估指標時，隨機森林模型在上述兩個製程中對異常偵測的表現，分別比決策樹學習提升了12.75%與6.86%。決策樹之視覺化結構顯示，「等待貼合時間」與「外觀重工數量」為決定平面膜層貼合製程品質之關鍵特徵，在軌跡式塗膠製程中，過度的人為修補行為會顯著的放大風險，造成品質不穩定，因為該製程對視覺對位公差和人為重工的累積疲勞相當敏感。本研究不僅建立了早期異常預警機制，更提供製造管理單位具體之風險閾值，最小化重工成本與提升製程穩定度之實務目標。

The bonding of optical components is a critical pre-processing step for determining the quality of electronic products. This step involves the cooperation of human and machines to satisfy the various needs from customers. Tracing the quality of individual item is infeasible to find abnormality in this kind of production processes. This study thus considers work orders as the level of analysis for early abnormal detection. The data collected from flat film lamination and trajectory-dispensing processes in two years showed that the imbalanced ratios for these two processes are 32.97 and 45.53, respectively. Oversampling method SMOTE-NC is first employed to balance the class distribution in a data set, and misclassification costs are then set for decision tree learning and random forest to improve the classification performance on abnormal work orders. When evaluation metric is F-measure, the experimental results showed that random forest improves anomaly detection by 12.75% and 6.86% over decision tree learning in the two processes. The decision tree for flat film lamination suggests that‘waiting time for bonding’and‘cosmetic reworks’are the most critical features. In trajectory-dispensing process, excessive manual rework significantly amplifies risks, because this process is sensitive to visual alignment tolerance and manual rework cumulative fatigue. This study not only establishes an early-warning mechanism, but also provides concrete risk thresholds for manufacturing management to achieve the goals of minimizing rework costs and enhancing process stability.

鍾佩真. (2022)。不平衡資料分類效能衡量指標之性質分析。國立成功大學資訊管理研究所碩士論文。
Akbani, R., Kwek, S., & Japkowicz, N. (2004). Applying support vector machines to imbalanced datasets. Proceedings of the 15th European Conference on Machine Learning, 39-50, Pisa, Italy.
Aly, M., & Behiry, M. H. (2025). Enhancing anomaly detection in IoT-driven factories using Logistic Boosting, Random Forest, and SVM: A comparative machine learning approach. Scientific Reports, 15(1), 23694.
Bader-El-Den, M., Teitei, E., & Perry, T. (2018). Biased random forest for dealing with the class imbalance problem. IEEE Transactions on Neural Networks and Learning Systems, 30(7), 2163-2172.
Bahadur, B., Sampica, J. D., Tchon, J. L., & Butterfield, A. (2011). Direct dry film optical bonding - a low-cost, robust, and scalable display lamination technology. Journal of the Society for Information Display, 19(11), 732-740.
Baldi, P., Brunak, S., Chauvin, Y., Andersen, C. A., & Nielsen, H. (2000). Assessing the accuracy of prediction algorithms for classification: an overview. Bioinformatics, 16(5), 412-424.
Blankenbach, K. (2016). What is a display? An introduction to visual displays and display systems. In Handbook of Visual Display Technology (pp 1-22). Springer International Publishing.
Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123-140.
Breiman, L. (2001). Random forests. Machine learning, 45(1), 5-32.
Brown, J. B. (2018). Classifiers and their metrics quantified. Molecular informatics, 37(1-2), 1700127.
Bunkhumpornpat, C., Sinapiromsaran, K., & Lursinsap, C. (2009). Safe-level-smote:safe-level synthetic minority over-sampling technique for handling the class imbalanced problem. Proceedings of the 13th Pacific-Asia Conference on Knowledge Discovery and Data Mining, 5476, 475-482, Bangkok, Thailand.
Cao, B., Liu, Y., Hou, C., Fan, J., Zheng, B., & Yin, J. (2020). Expediting the accuracy-improving process of SVMs for class imbalance learning. IEEE Transactions on Knowledge and Data Engineering, 33(11), 3550-3567.
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). Smote: synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16, 321-357.
Chen, W., Yang K., Yu, Z., Shi, Y., & Chen, C. P. (2024). A survey on imbalanced learning: latest research, applications and future directions. Artificial Intelligence Review, 57(6), 1-51.
Cruz, S., Sousa, A., Viana, J. C., & Martins, T. (2017). Analysis of the bonding process and materials optimization for mitigating the yellow border defect on optically bonded automotive display panels. Displays, 48, 21-28.
Datta, S., Ghosh, A., Sanyal, K., & Das, S. (2017). A radial boundary intersection aided interior point method for multi-objective optimization. Information Sciences, 377, 1-16.
Dighde, R. M., Schultz, B., & Keam, N. (2016) Display device. U.S. Patent No. 9383769 B2.
Douzas, G., Bacao, F., & Last F. (2018). Improving imbalanced learning through a heuristic oversampling method based on k-means and smote. Information Sciences, 465, 1-20.
Elkan, C. (2001). The foundations of cost-sensitive learning. In International joint conference on artificial intelligence, 17(1), 973-978. Lawrence Erlbaum Associates Ltd.
Fawcett, T. (2006). An introduction to ROC analysis. Pattern Recognition Letters, 27(8), 861-874.
Freund, Y. & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences, 55(1), 119-139.
Gu, Q., Zhu, L., & Cai, Z. (2009). Evaluation measures of the classification performance of imbalanced data sets. Proceedings of the International Symposium on Intelligence Computation and Applications, 461-471, Springer, Berlin, Heidelberg. 
Han, H., Wang, W. Y., & Mao, B. H. (2005). Borderline-smote: a new over-sampling method in imbalanced datasets learning. International Conference on Intelligent Computing, 3644, 878-887, Hefei, China.
Kim, S., An, J., Son, S. R., Choi, J. W., Park, J., Park, C. B., & Lee, J. H. (2022). Fabrication of highly elastic and optically transparent adhesive films for flexible displays using multifunctional photocrosslinker. Molecular Crystals and Liquid Crystals, 740(1), 28-34.
Kim, Y. G., Park B. M., & Han, K. Y. (2022) Optimization of lamination process for flexible display using OCR. Molecular Crystals and Liquid Crystals, 734(1), 35-46.
Li, L., He, H., & Li, J. (2019). Entropy-based sampling approaches for multi-class imbalanced problems. IEEE Transactions on Knowledge and Data Engineering, 32(11), 2159-2170.
Mahale, Y., Kolhar, S., & More, A. S. (2025). Enhancing predictive maintenance in automotive industry: addressing class imbalance using advanced machine learning techniques. Discover Applied Sciences, 7(4), 340.
Matthews, B. W. (1975). Comparison of the predicted and observed secondary structure of T4 phage lysozyme. Biochimica et Biophysica Acta (BBA)-Protein Structure, 405(2), 442-451.
Oliveira, R., Viana, J., & Sampaio P. (2023). Optical bonding process of flat panel displays and their critical to quality factors. The International Journal of Advanced Manufacturing Technology, 125(11), 5631-5650.
Pan, T., Zhao, J., Wu, W., & Yang, J. (2020). Learning imbalanced datasets based on SMOTE and Gaussian distribution. Information Sciences, 512, 1214-1233.
Sáez, J. A., Luengo, J., Stefanowski J., & Herrera F. (2015). Smote-ipf: addressing the noisy and borderline examples problem in imbalanced classification by a re-sampling method with filtering. Information Sciences, 291, 184-203.
Shamrat, F. J. M., Ranjan, R., Md, K., Hasib, A. Y., & Siddique, A. H. (2022). Performance evaluation among id3, c4. 5, and cart decision tree algorithm. Proceedings of International Conference on Pervasive Computing and Social Networking 2021, 317, 127–142, Salem, India.
Sun, Y., Wong, A. K., & Kamel, M. S. (2009). Classification of imbalanced data: a review. International Journal of Pattern Recognition and Artificial Intelligence, 23(4), 687-719.
Witten, I. H., Frank, E., Hall, M. A., & Pal, C. J. (2017). Practical machine learning tools and techniques. In Data Mining (pp. 403-413). Elsevier Publishers.
Yeh, G., Rudolph, M. L., & Chen, J. Y. C. (2013). Minimizing the Impact of Bonding Induced LCD Defects. In SID Symposium Digest of Technical Papers, 44(1), 951-954.