ScagCNN: Estimating Visual Characterizations of 2D Scatterplots via Convolution Neural Network

Scagnostics is a set of visual features that characterizes the data distribution of a 2D scatterplot and has been used in a wide range of applications. However, calculating the scagnostics scores involves computationally expensive algorithms. Moreover, the algorithms are sensitive to the slight changes in the underlying data distribution within the scatterplot. Therefore, this work provides a machine learning model, called ScagCNN, to estimate the scagnostics scores. This model aims to improve the scagnostics computation time and to reduce the sensitivity to the small shifts in the data distribution. This work also provides a web prototype to explore the predictive performance of the model and to give a visual explanation about whether a prediction is accurate. Furthermore, we test the performance of our solution on datasets of various sizes.