A Method for Optimizing the Rendering Order of Scatterplots

Rendering order is crucial for generating effective scatterplots, as a rendering sequence can cause anomalous data points to be obscured by others. This issue is particularly significant in the field of explainable artificial intelligence (XAI), where large volumes of data can prevent users from observing misclassified instances. This poster introduces a novel method for sorting data points and rendering them sequentially to reduce the likelihood of anomalous points being hidden. First, we normalize the two coordinates of the scatterplots to mitigate the impact of differing value ranges. Next, we propose a method for calculating the anomaly index of each data point. Finally, we sort the data points based on their anomaly index and render them sequentially.We compare our method with existing approaches on scatterplots generated by dimensionality reduction (DR) techniques applied to a pretrained convolutional neural network (CNN) trained on the MNIST dataset. The results demonstrate that our method enables easier identification of misclassified (anomalous) data points compared to category-based and random rendering orders.