This repository is a fork of original LEAF repository. The only adjustments are removing black-box model training and adjusting few data shapes, so that this framework works with our model and dataset.

Framework and implementation were suggested in: Amparore, E., Perotti, A., & Bajardi, P. (2021). To trust or not to trust an explanation: using LEAF to evaluate local linear XAI methods. PeerJ Comput. Sci., 7(6), e479. doi: 10.7717/peerj-cs.479

LEAF

A Python framework for the quantitative evaluation of eXplainable AI methods.

LEAF requires the following Python libraries to work:

numpy, pandas, lime, shap, imblearn, tabulate

The LEAF project directory contains the following files:

• leaf.py: the main code of LEAF, with the evaluation procedures for the LLE explaners LIME and SHAP
• LEAF_test.ipynb: a Jupyter notebook with a simple example of how to use LEAF to compute the basic metrics for XAI evaluation
• heartrisk-dataset.txt: the sample HeartRisk dataset provided by Shivakumar Doddamani, in csv format. See www.kaggle.com/shivakumarcd/heart-risk-problem for reference.