A Machine Learning Model Selection Considering Tradeoffs between Accuracy and Interpretability

Zhumakhan Nazir, Temirlan Zarymkanov, Jurn-Guy Park

2022

Full text Open access: Gold Metadata (CSL-JSON)

Data confidence 70 medium

How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Applying black-box ML models in high-stakes fields like criminology, healthcare and realtime operating systems might create issues because of poor interpretability and complexity.Also, model building methods that include interpretability is now one of the growing research topics due to the absence of interpretability metrics that are both model-agnostic and quantitative.This paper introduces model selection methods with trade off between interpretability and accuracy of a model.Our results show 97% improvement in interpretability with 2.5% drop in accuracy in AutoMPG dataset using MLP model (65% improvement in interpretability with 1.5% drop in accuracy in MNIST dataset).