Explainable AI: understand the black box of predictive models with chemical engineering applications

This project will focus on the explainability of machine learning predictive models for drop coalescence conducted in microfludics device and use initial experimental conditions as input. In the first stage, the student will construct different ML predictive models, including random forest, XGboost and deep neural networks. The second stage consists of extensive analysis of model results/parameters. SHAP and LIME will be used for model interpretbility.

Getting Started

Programming language: Python 3.8.16

Software Requirement

Environment Installation

First of all, please set up a conda environment

conda env create -n XAI 
conda activate XAI
pip install numpy pandas scikit-learn xgboost matplotlib seaborn shap tensorflow lime

Package Versions

Package Name	Version
_py-xgboost-mutex	2.0
libxgboost	1.7.4
lime	0.2.0.1
matplotlib-base	3.7.1
matplotlib-inline	0.1.6
numpy	1.24.3
pandas	2.0.1
py-xgboost	1.7.4
scikit-learn	1.2.2
seaborn	0.12.2
shap	0.41.0
tensorflow	2.13.0rc1
tensorflow-estimator	2.13.0rc0
tensorflow-macos	2.13.0rc1
xgboost	1.7.4

Dataset

We use min-max normlazation to rescale the raw dataset, and split it into two for training and testing repectively. Validation step is evaluated by using 5-fold cross validation. The experimental data can be available upon reasonable request to Dr. Nina Kovalchuk (n.kovalchuk@bham.ac.uk)

	Coalescence	Non-Coalescence	Total
Total Dataset	782	719	1501
Training Dataset	625	575	1200
Testing Dataset	157	144	301

Implementation

Generating Predictive Results

We trains random forest, XGBoost and MLP classifier to get predictions. The corresponding implementation code are shown in code folder.

Interpret predictive models

We use SHAP and LIME to get model interpetbility. Feature ablation testing is also used to analyse feature importance. The feature ablation results is shown in code/ablation_test.ipynb