Simple Interaction Finding Technique (SIFT) – A Simple Methodology to Generate Novel Hypotheses from Complex Datasets

Authors

Murtaza Nasir, University of Massachusetts LowellFollow
Nichalin Summerfield, University of Massachusetts LowellFollow
Serhat Simsek, Montclair UniversityFollow
Asil Oztekin, University of Massachusetts LowellFollow

Abstract

Machine Learning (ML) models have become ubiquitous in all spheres of research and decision-making. Understanding ML models as well as the data-generation-process (DGP) for the dataset under examination are important. Most highly accurate ML models are blackboxes that aren't interpretable. In this work, we propose a methodology that can help elicit important information from any ML models. Our methodology allows the use of any highly-accurate ML model to find interactions between variables in the dataset. This can allow for a better understanding of the underlying DGP by using a data-and-model agnostic process to synthesize new knowledge about the underlying phenomenon.

Recommended Citation

Nasir, Murtaza; Summerfield, Nichalin; Simsek, Serhat; and Oztekin, Asil, "Simple Interaction Finding Technique (SIFT) – A Simple Methodology to Generate Novel Hypotheses from Complex Datasets" (2021). NEAIS 2021 Proceedings. 6.
https://aisel.aisnet.org/neais2021/6