MUCH

MUlti Counterfactual Halton sampling

MUCH

MUlti Counterfactual Halton sampling

Counterfactual eXplanations (CEX) are spreading rapidly in the literature on explainable AI due to their ability to be easily understood and highly adaptable to data. In this repository you will found an original method for the extraction of multiple CEX from numerical data based on Halton sampling and Support Vector Data Description.

Examples may help to better understand the motivation behind this study and the importance of counterfactual reasoning in multiclass problems. In the field of health, several diseases present with different stages of severity (e.g., cancer, chronic obstructive pulmonary disease...) that can worsen drastically in a short time if not properly treated. In this case, multiclass counterfactuals can be a crucial instrument to monitor the stage of disease progression in order to detect minimal changes in the patient’s condition and apply appropriate countermeasures before the disease progresses to the next stage. Another example may instead involve the study of the transition of a phenomenon that develops over several stages (e.g., A, B, C, D). A counterfactual analysis can be useful to check for differences between different transitions (e.g., direct paths skipping intermediate transitions or progressive sequential paths).

The concept paper behind this algorithm is currently under review at the IEEE Transactions on Artificial Intelligence.

Enjoy !

Requirements

For the evaluation and visualization of the counterfactual explanations we used the following github repository:

Moses (2023). spider_plot (https://github.com/NewGuy012/spider_plot/releases/tag/19.4), GitHub. Retrieved January 17, 2023.

The datasets can be directly retrieved from Kaggle_

-) FIFA 2023: https://www.kaggle.com/datasets/cashncarry/fifa-23-complete-player-dataset

-) Stellar Classification: https://www.kaggle.com/fedesoriano/stellar-classification-dataset-sdss17