Using Shape Metrics to Describe 2D Data Points
This addresses the need for more objective and interpretable model building, particularly in medical applications where explainability is crucial, though it appears incremental as it builds on existing shape metric concepts.
The paper tackles the problem of subjectivity and poor reproducibility in traditional machine learning by proposing the use of shape metrics to describe 2D data, aiming to automate aspects of model building for improved explainability and interpretability.
Traditional machine learning (ML) algorithms, such as multiple regression, require human analysts to make decisions on how to treat the data. These decisions can make the model building process subjective and difficult to replicate for those who did not build the model. Deep learning approaches benefit by allowing the model to learn what features are important once the human analyst builds the architecture. Thus, a method for automating certain human decisions for traditional ML modeling would help to improve the reproducibility and remove subjective aspects of the model building process. To that end, we propose to use shape metrics to describe 2D data to help make analyses more explainable and interpretable. The proposed approach provides a foundation to help automate various aspects of model building in an interpretable and explainable fashion. This is particularly important in applications in the medical community where the `right to explainability' is crucial. We provide various simulated data sets ranging from probability distributions, functions, and model quality control checks (such as QQ-Plots and residual analyses from ordinary least squares) to showcase the breadth of this approach.