Automated XAI Pipeline for Feature Attribution and Functional Similarity Analysis in Experimental Solid Foam Data - Dataset, Codes, and Results

The repository provides all code and experimental data required to reproduce the automated two‑stage XAI pipeline for feature attribution and functional similarity analysis proposed in the paper “Automated XAI Pipeline for Feature Attribution and Functional Similarity Analysis in Experimental Solid Foam Data,” currently under review. It contains the solid‑foam dataset with nineteen foams of different materials, their structural descriptors, operating conditions, and measured functional responses for two use cases: pressure drop per unit length and average heat‑transfer coefficient under varying gas flow rates. This Zenodo repository includes: Curated tabular datasets for both use cases; Scripts for the regression stage, where Optuna‑tuned MLP regressors are trained and analyzed using SHAP and permutation feature importance; and Scripts for the classifier‑based similarity stage, where MLP classifiers are trained on functional response pairs and final‑layer weights are used to derive foam‑to‑foam similarity matrices. It also provides configuration files, result artifacts (e.g., parameters of trained models, similarity matrices, overlap statistics), and documentation of the physics‑informed validation protocol (code) that compares learned similarity neighborhoods with polynomial‑fit reference neighborhoods, enabling transparent reproduction and extension of the study to other experimental material systems with functional response data. For further information, please consult the README.md file.