<p>The classification of defect patterns in wafer maps is a crucial task in the semiconductor industry. By accurately identifying and classifying these patterns, manufacturers can take necessary measures to improve the quality of their wafers and optimize their production processes. To address this challenge, we propose a stacking ensemble method and utilizes data augmentation. In the first step, to address the issue of data imbalance resulting from difficulties in collecting data in real-world environments, we apply data augmentation techniques with convolutional autoencoder to expand the training dataset. This helps to minimize the impact of minority sample class and improve the generalization ability of the classification model. In the second step, we leverage the stacking ensemble method, which combines multiple classification models to make more accurate predictions. By combining handcrafted features with deep features in training deep convolutional models, this approach has been shown to improve the classification performance and provide valuable support for fault diagnosis in the semiconductor industry. Experimental evaluation is performed with stacking models with fivefold cross-validation, evaluating and comparing them in terms of accuracy, precision, recall, and F1-score. The proposed method achieves an average classification accuracy of 0.9818, a micro-F1-score of 0.9818, and a macro-F1-score of 0.9318.</p>