Data Selection: A General Principle for Building Small Interpretable Models

TL;DR

This paper demonstrates that sampling training data based on learned data distribution significantly improves the accuracy of small, interpretable models across various tasks, making them competitive with modern techniques.

Contribution

It provides a rigorous validation of data sampling from learned distributions as a general strategy for enhancing small model accuracy in diverse practical settings.

Findings

Improved accuracy of cluster explanation trees.

Enhanced prototype-based classification performance.

Effective use of data sampling with Random Forests.

Abstract

We present convincing empirical evidence for an effective and general strategy for building accurate small models. Such models are attractive for interpretability and also find use in resource-constrained environments. The strategy is to learn the training distribution and sample accordingly from the provided training data. The distribution learning algorithm is not a contribution of this work; our contribution is a rigorous demonstration of the broad utility of this strategy in various practical settings. We apply it to the tasks of (1) building cluster explanation trees, (2) prototype-based classification, and (3) classification using Random Forests, and show that it improves the accuracy of decades-old weak traditional baselines to be competitive with specialized modern techniques. This strategy is also versatile wrt the notion of model size. In the first two tasks, model size is considered to be number of leaves in the tree and the number of prototypes respectively. In the final task involving Random Forests, the strategy is shown to be effective even when model size comprises of more than one factor: number of trees and their maximum depth. Positive results using multiple datasets are presented that are shown to be statistically significant.