- https://github.com/hwd404/FOLD-R-PP (FOLD-R++ algorithm GitHub repo)
- https://arxiv.org/abs/2110.07843 (Research paper describing the FOLD-R++ algorithm by UTD's Dr. Gopal Gupta)
My classification has an accuracy of 100% on a test data set comprised of 20% of the original data set, with the other 80% devoted to training. I generated an explainable answer set programming (ASP) rule set using the FOLD-R++ algorithm. The benefit of this method is that it creates a human-comprehensible rule set to determine if a mushroom is poisonous or not, avoiding the black-box model of traditional machine learning classification.
One can see my code and output in the coding-challenge.ipynb file, which is a Jupyter Notebook. The code consists of setting up the model and importing and processing the data and the use of the model. The generated answer set programming rule set achieved 100% accuracy on the test data using only 4 default rules and 3 exception rules. Finally, in the last output, one can see the explanation for the classification of each mushroom in the test data set.
A concrete example:
Explanation for example number 37:
answer 1:
[T]class(X,'p') :- not [F]gill-size(X,'b'), [T]stalk-surface-below-ring(X,'y').
This states that test mushroom number 37 is poisonous because its gill size is not broad and its stalk surface below its ring is scaly. The model learned these rules only from the training data and the FOLD-R++ algorithm.
The FOLD-R++ algorithm is very powerful since it gives explanations to its predictions in addition to being very accurate, especially on this data set. The rules are simple, logical, and understandable and still bring amazing predictive power. I would like to thank Dr. Gopal Gupta for showing this technology off at a workshop for HackReason 2022 and highlighting the importance and practicality of explainable AI. Explainable AI is important for transparency and the ability to correct biases and incorrect or ungeneralizable assumptions.