Reference tables to introduce and organize evaluation methods and measures for explainable machine learning systems

[NeurIPS 2024] CoSy is an automatic evaluation framework for textual explanations of neurons.

Replication package for the KNOSYS paper titled "An Objective Metric for Explainable AI: How and Why to Estimate the Degree of Explainability".

Open and extensible benchmark for XAI methods

Semantic Meaningfulness: Evaluating counterfactual approaches for real world plausibility

Saliency Metrics is a Python package that implements various metrics for comparing saliency maps generated by explanation methods.

Repository for ReVel framework to Measure Local-Linear Explanationsfor Black-Box Models

CNN architectures Resnet-50 and InceptionV3 have been used to detect whether the CT scan images is covid affected or not and prediction is validated using explainable AI frameworks LIME and GradCAM.

ConsisXAI is an implementation of a technique to evaluate global machine learning explainability (XAI) methods based on feature subset consistency

A dual-headed deep learning model built using TensorFlow and Keras to classify fruit type (Apple, Banana, Guava, Orange) and their quality condition (Good or Bad) from images. The system includes Grad-CAM-based visual explanations and a responsive Streamlit web interface for real-time predictions using uploaded images or webcam input.

Code for evaluating saliency maps with classification metrics.

Research on AutoML and Explainability．

This project poses a new methodology for assessing and improving sequential concept bottleneck models (CBMs). The research undertaken in this project builds upon the model proposed by Grange et al., of which I was one of the co-authors.

A step by step tutorial on the Temporal Fusion Transformer for time series forecasting using PyTorch. The project uses daily cryptocurrency data obtained from Yahoo Finance and includes a custom function designed to ensure full reproducibility of experimental results.

A framework for evaluating explainable AI (XAI) methods in drug discovery using multiple machine learning architectures. This repository implements three distinct model architectures (CNN, Random Forest, and RGCN) and provides a hierarchical four-tier evaluation framework for assessing the quality and reliability of their explanations.

This repository is the code basis for the paper titled "Balancing Privacy and Explainability in Federated Learning"

Working prototype for semantic AAC pictogram generation

Scripts and trained models from our paper: M. Ntrougkas, V. Mezaris, I. Patras, "P-TAME: Explain Any Image Classifier with Trained Perturbations", IEEE Open Journal of Signal Processing, 2025. DOI:10.1109/OJSP.2025.3568756.

Classify applications using flow features with Random Forest and K-Nearest Neighbor classifiers. Explore augmentation techniques like oversampling, SMOTE, BorderlineSMOTE, and ADASYN for better handling of underrepresented classes. Measure classifier effectiveness for different sampling techniques using accuracy, precision, recall, and F1-score.