Data-driven fault classification is critical for wind turbine condition monitoring systems. However, imbalanced class distributions, where certain fault types are significantly less frequent than others, pose a challenge for traditional classifiers. This work addresses this issue by investigating the combined effect of oversampling with Synthetic Minority Oversampling Technique (SMOTE) and undersampling with Tomek Links for data cleaning, followed by a stacking ensemble learning approach. We evaluate the proposed method on a multiclass wind turbine fault dataset and demonstrate its effectiveness in improving classification performance compared to baseline methods. Our findings highlight the potential of the proposed approach for enhancing fault classification accuracy in wind turbine systems plagued by imbalanced data. The classification performance is evaluated using metrics like confusion matrix, precision, recall, and F1-score, demonstrating significant improvements compared to baseline methods.