A Machine Learning Specialist is building an image classification model for identifying car brands. The image data used for training the model consists of two classes namely Brand X and Brand Y. After reaching the global minimum, the Specialist immediately conducted tests on new data, however, the model performance was not acceptable. The Specialist has observed that 80% of the misclassified images are that of Brand Y facing upside down. Which strategy would MOST likely produce the best output for the model?