IMPTOX - WP1 - D1.9 Data Analysis Report Appendix

This repository is confidential and only intended for member of the IMPTOX consortium. The IMPTOX project has received funding from the EU's H2020 framework programme for research and innovation under grant agreement n. 965173. Imptox is part of the European MNP cluster on human health. More information about the project here. Archive content This repository contains the pdf appendix and report of D1.9 as well as Data (csv, plots) and code (Python Notebooks). Datasets The datasets folder contains the RAW data as well as the processed data that are used for further analysis. The data preprocessing is described in the notebook as well as in the report (D1.9) in the Work Packages section of IMPTOX [intranet](https://intranet.imptox.eu/WorkPackages#home) or in this archive. The original data were not generated by WP1. Notebooks Five notebooks are presented here: 0_data.ipynb : takes care of the data preprocessing and train/test subset creation. 1_FuzzyCoCo.ipynb : train the FuzzyCoCo model and explore various fuzzy systems and human readable rules. 2_SHAP.ipynb : Trains a Multi Layer Perceptron and extracts the features weights of the predictions for each feature with SHAP. 3_GBoost.ipynb : Identical as 2 but with Gradient Boosting. 4_reporting.ipynb : Creates general plots used in the report. Each notebook creates various plots as well as intermediate data that are saved as CSVs in the corresponding subfolders. Note that as we train our models on eight different output variables, the exported data (plots and csv) are present eight times for each method. The exported files include: SHAP values Model performance metrics Data train/test split Saved Models Plots Usage note To run the notebooks, a classic anaconda virtual environment with basic datascience packages is sufficient in most cases. For specific tools, such as Fuzzy CoCo, additional informations are available within the concerned notebooks. Supplementary material for eXplainable Artificial Intelligence Regarding the various XAI methods used in this task, you can go to this GitHub.io page where the different methods and more are described in details with additional external ressources.