IDEAS SoCal AI & Data Science Conference 2019

Applications of machine learning in commercial, educational, and even judicial settings are increasingly subject to regulation and to scrutiny for adverse effects such as biases and discrimination. For example, the US Equal Credit Opportunity Act requires creditors to provide an explanation to an applicant when their models recommend rejecting the applicant for credit, and such explanations require model interpretability. The controversy around discrimination in algorithmic decisions have further highlighted the need for transparent models whose recommendations can be understood, in contrast to black-box algorithms where the step from input data to output decision is opaque.

With these challenges of explainability in mind, we present S3D, a machine learning model that is both fully interpretable and highly performant at prediction, along with its open-source Python package. This model is easily analyzable and indeed visualizable, giving researchers and practitioners a method of visually inspecting their data and the relationship between the outcome variable of interest and the predictive features or covariates. S3D can predict outcomes in the held-out data at levels comparable to state-of-the-art approaches, but in addition, produces interpretable models that allow for explainable predictions.

S3D can be a high utility interpretable modelling tool for both the scientific and commercial domains, and in the presentation, we demonstrate the Python S3D package and its capabilities for prediction, feature selection, feature clustering, and explainability. We will demo how to get up and running with the package, and illustrate through examples of its application to real datasets.