Interpretable Clothing Size Prediction: CNN and XAI for Personalized Recommendations

This research work proposes a clothing size prediction method based on customer images using Convolutional Neural Networks (CNN) and incorporating Explainable Artificial Intelligence (XAI) methods. The proposed CNN model directly computes size recommendations based on typical key visual attributes about body dimensions and fitting preferences of clothing. Both training and testing were performed with a set of images of labeled customers; the overall accuracy of the model reached 89%; precision and recall were higher than 90% for the majority of size categories. The blend of XAI enabled the visualization and even the explanation of the modality to show how any particular feature of the visual managed to portray such a conclusion. It also makes the model highly interpretable and thus very reliable for e-commerce applications, where accurate size predictions and recommendations are especially important. It is possible to state that the proposed method can promisingly improve the shopping experience through more accurate and individualized forecasts of the size of the clothes, minimizing high rates of return and thereby increasing customer satisfaction.