FOLD-SE: An Efficient Rule-based Machine Learning Algorithm with Scalable Explainability

Explore an efficient, explainable machine learning algorithm for classification tasks called FOLD-SE in this 37-minute conference talk presented at PADL'24. Discover how FOLD-SE generates a set of default rules as an explainable model for tabular data containing numerical and categorical values. Learn about the novel Magic Gini Impurity heuristic for literal selection, the refined data comparison operator, and how FOLD-SE eliminates the long tail effect. Understand the scalable explainability provided by FOLD-SE, which maintains a small number of learned rules and literals regardless of dataset size while preserving good classification accuracy. Compare FOLD-SE's competitive performance against traditional machine learning algorithms like XGBoost and Multi-Layer Perceptrons, noting its order of magnitude faster speed and explainable model generation. Examine how FOLD-SE outperforms prior rule-learning algorithms such as RIPPER in efficiency, performance, and scalability, particularly for large datasets, while generating significantly fewer rules than earlier default rule learning algorithms like FOLD-R++.

[PADL'24] FOLD-SE: An Efficient Rule-based Machine Learning Algorithm with Scalable Explai...