Enhancing Fashion Trend Predictions with XAICNN

This study analyzes the Explainable AI-enabled Convolutional Neural Network (XAI-CNN) model for image trend analysis and classification. It is mainly aimed at increasing the interpretability and prediction performance of model predictions. We built two XAI-CNN models embedding Saliency Maps and Grad-CAM to explain our model visually. As a result of these experiments, the XAI-CNN models show markedly higher accuracy than a Baseline CNN model. More specifically, the Baseline CNN and XAI-CNN model with Saliency Maps had an accuracy of 85% only while the one using Grad-CAM achieved a top performance touching on almost $\mathbf{9 0 \%}$. These findings emphasize the importance of using explainable AI methods to improve model accuracy as well as interpretability. The visual explanations provided by the XAICNN models reveal important knowledge regarding what contributes to fashion trends, flowing back to related industries and stakeholders to assist their decisions. This study shows the relevance of BI in AI models as a whole, especially those where interpretability becomes one urgent requirement. Further investigations are recommended that can use other XAI techniques and multimodal data to improve the prediction performance of IPR-HRCT APWH while providing more interpretable results.