Case Studies
Preface
Technical Setup
1
Explainable Artificial Intelligence 1
1.1
Explaining Credit Card Customers churns
1.1.1
Introduction and Motivation
1.1.2
Methodology
1.1.3
Local explanations
1.1.4
Global explanations
1.1.5
Summary
1.2
XAI in real estate pricing: A case study
1.2.1
Abstract
1.2.2
Introduction
1.2.3
Related Work
1.2.4
Methodology
1.2.5
Results
1.2.6
Summary and conclusions
1.3
Coronary artery disease: Is it worth trusting ML when it comes to our health?
1.3.1
Abstract
1.3.2
Introduction and Motivation
1.3.3
Related Work
1.3.4
Methodology
1.3.5
Results
1.3.6
Summary and conclusions
1.4
Red wine mystery: using explainable AI to inspect factors behind wine quality
1.4.1
Abstract
1.4.2
Introduction and Motivation
1.4.3
Methodology
1.4.4
Global explanations
1.4.5
Local explanations
1.4.6
Confrontation with science
1.4.7
Summary
1.4.8
Conclusions
1.5
Explanatory approach to modeling the risk of hotel booking cancellations
1.5.1
Abstract
1.5.2
Introduction
1.5.3
Dataset and models
1.5.4
Local explanations
1.5.5
Global explanations
1.5.6
Summary and conclusions
2
Explainable Artificial Intelligence 2
2.1
Does brand has an impact on smartphone prices?
2.1.1
Abstract
2.1.2
Introduction, motivation and related work
2.1.3
Methodology
2.1.4
Results
2.1.5
Summary and conclusion
2.2
Classifying people as good or bad credit risks
2.2.1
Abstract
2.2.2
Introduction
2.2.3
Dataset and models
2.2.4
Local explanations
2.2.5
Global explanations
2.2.6
Summary and conclusions
2.3
How to predict the probability of subsequent blood donations?
2.3.1
Abstract
2.3.2
Introduction and motivation
2.3.3
Related work
2.3.4
Data analysis and pre-processing
2.3.5
Models preparation and selection
2.3.6
Global explanations
2.3.7
Local explanations
2.3.8
Conclusions and summary
2.4
Explaining diabetes indicators
2.4.1
Abstract
2.4.2
Introduction
2.4.3
Methods
2.4.4
Explanations
2.4.5
Results
2.4.6
Expert opinions
2.4.7
Conclusions
2.5
How the price of the house is influenced by neighborhood? XAI methods for interpretation the black box model
2.5.1
Abstract
2.5.2
Introduction
2.5.3
Literature
2.5.4
Local explanations
2.5.5
Global explanations
2.5.6
Conclusion
3
Deep Learning 1
3.1
LungNet
3.1.1
Introduction
3.1.2
Data
3.1.3
Original model
3.1.4
New models
3.1.5
Summary
3.2
On the reproducibility of the BCDU-Net model
3.2.1
Abstract
3.2.2
Introduction
3.2.3
Reproduction of the results
3.2.4
Further experiments
3.2.5
Other tools applied to the model
3.2.6
Results and conclusions
3.3
An Exploration of DeepCovidExplainer: Explainable COVID-19 Diagnosis from Chest X-rays
3.3.1
Introduction and motivation
3.3.2
Related work
3.3.3
Our work
3.3.4
Conclusions and summary
3.4
ERSCovid
3.4.1
Abstract
3.4.2
Problem
3.4.3
ERSCovid workflow
3.4.4
Data
3.4.5
Training
3.4.6
Changes to original model
3.4.7
Summary
3.5
Reproducing and modifying COVID-Net
3.5.1
Abstract
3.5.2
Introduction and motivation
3.5.3
Related work
3.5.4
Dataset
3.5.5
Reproducibility
3.5.6
Further experiments and network modifications
3.5.7
Summary
4
Deep Learning 2
4.1
What makes an article reproducible? Comparison of the FER+ paper and AxonDeepSeg
4.1.1
Abstract
4.1.2
Introduction
4.1.3
Analyzing the FERPlus paper
4.1.4
Reproducibility analysis
4.1.5
Analyzing the AxonDeepSeg paper
4.1.6
Reproducibility analysis
4.1.7
Conclusion
4.2
Can you classify histopathological data at home? Reproducing the ARA-CNN model’s data and performance.
4.2.1
Abstract
4.2.2
Introduction
4.2.3
Definition
4.2.4
Methodology
4.2.5
Results
4.2.6
Summary and conclusions
4.3
Rethinking the U-Net architecture for multimodal biomedical image segmentation
4.3.1
Abstrac
4.3.2
What Reproducibility Is?
4.3.3
First article (An Improvement of Data Classification Using Random Multimodel Deep Learning (RMDL) )
4.3.4
Second article (MultiResUNet : Rethinking the U-Net architecture for multimodal biomedical image segmentation)
4.3.5
Conclusion
4.4
The reproducibility analysis of articles covering RMDL and UNet++ architectures churns
4.4.1
Abstract
4.4.2
Reproduction
4.4.3
Random Multimodel Deep Learning for Classification
4.4.4
A Nested U-Net Architecture for Medical Image Segmentation
4.4.5
Conclusions
4.5
Can you trust science? On Reproduction in Deep Learning.
4.5.1
Abstract
4.5.2
Introduction
4.5.3
Methodology
4.5.4
Result
4.5.5
Discussion
5
Machine Learning
5.1
Validation and comparison of COVID-19 mortality prediction models on multi-source data
5.1.1
Abstract
5.1.2
Introduction
5.1.3
Data description
5.1.4
Comparison of the models
5.1.5
Conclusion
5.2
One model to fit them all: COVID-19 survival prediction using multinational data
5.2.1
Abstract
5.2.2
Introduction
5.2.3
Data sources
5.2.4
Model building
5.2.5
Discussion
5.2.6
Summary
5.3
Transparent machine learning to support predicting COVID-19 infection risk based on chronic diseases
5.3.1
Abstract
5.3.2
Introduction
5.3.3
Flaws
5.3.4
Improvements
5.3.5
Transparent Machine Learning
5.3.6
Application
5.3.7
Conclusions
5.4
Comparison of neural networks and tree-based models in the clinical prediction of the course of COVID-19 illness
5.4.1
Abstract
5.4.2
Introduction
5.4.3
Methods
5.4.4
Results
5.4.5
Discussion
5.4.6
Source code
6
RashomonML
6.1
How to compare many good machine learning models?
6.1.1
Abstract
6.1.2
Introduction
6.1.3
Literature review
6.1.4
Methodology
6.1.5
Results
6.1.6
Summary and conclusions
6.2
Analysis of models predicting death probability during ICU stays
6.2.1
Abstract
6.2.2
Introduction
6.2.3
Literature review
6.2.4
Methodology
6.2.5
Results
6.2.6
Summary
6.3
Rashomon ML with addition of dimensional reduction
6.3.1
Abstract
6.3.2
Introduction and related works
6.3.3
Methodology
6.3.4
Results
6.3.5
Summary and conclusions
6.4
Rashomon sets on death prediction XGB models using MIMIC-III database
6.4.1
Abstract
6.4.2
Related works and introduction
6.4.3
Methodology
6.4.4
Results
6.4.5
Conclusions
6.5
Rashomon sets of in-hospital mortality prediction random forest models
6.5.1
Abstract
6.5.2
Introduction
6.5.3
Related work
6.5.4
MIMIC-III Dataset
6.5.5
Rashomon Sets
6.5.6
Results
6.5.7
Conclusion
Published with bookdown
Case Studies
Chapter 2
Explainable Artificial Intelligence 2