Machine Learning and Knowledge Extraction: 6th IFIP TC 5, TC 12, WG 8.4, WG 8.9, WG 12.9 International Cross-Domain Conference, CD-MAKE 2022, Vienna, ... (Lecture Notes in Computer Science, 13480)

This book constitutes the refereed proceedings of the 6th IFIP TC 5, TC 12, WG 8.4, WG 8.9, WG 12.9 International Cross-Domain Conference, CD-MAKE 2022, held in Vienna, Austria during August 2022. The 23 full papers presented were carefully reviewed and selected from 45 submissions. The papers are covering a wide range from integrative machine learning approach, considering the importance of data science and visualization for the algorithmic pipeline with a strong emphasis on privacy, data protection, safety and security. Preface Organization Contents Explain to Not Forget: Defending Against Catastrophic Forgetting with XAI 1 Introduction 2 Related Work 2.1 Explainable Artificial Intelligence 2.2 Catastrophic Forgetting 3 Relevance-Based Neural Freezing 4 Experiments 4.1 MNIST-Split 4.2 MNIST-Permuted 4.3 CIFAR10 and CIFAR100 4.4 ImageNet Split 4.5 Adience 4.6 Qualitative Results 5 Conclusion A Appendix A.1 MNIST-Split A.2 MNIST-Permuted A.3 CIFAR10 and CIFAR100 A.4 ImageNet Split A.5 Adience References Approximation of SHAP Values for Randomized Tree Ensembles 1 Interpreting Model Predictions 1.1 Conditional Feature Contributions (CFCs) 1.2 SHAP Values 2 SHAP Versus CFCs 2.1 Data Investigated 2.2 Comparative Study: Local Explanations 2.3 Predictive Power of Feature Subsets 3 Discussion References Color Shadows (Part I): Exploratory Usability Evaluation of Activation Maps in Radiological Machine Learning 1 Introduction 2 Methods 2.1 Data 2.2 Model Development 2.3 Activation Maps 2.4 Utility and Other Subjective Dimensions 3 Results 4 Discussion 5 Conclusion References Effects of Fairness and Explanation on Trust in Ethical AI 1 Introduction 2 Related Work 2.1 AI Fairness and Trust 2.2 AI Explanation and Trust 3 Method 3.1 Case Study 3.2 Explanations 3.3 Fairness 3.4 Task Design 3.5 Trust Scales 3.6 Experiment Setup 3.7 Participants and Data Collection 4 Results 4.1 Effects of Fairness on Trust 4.2 Effects of Explanation on Trust 4.3 Effects of Fairness and Explanation on Trust 5 Discussion 6 Conclusion and Future Work References Towards Refined Classifications Driven by SHAP Explanations 1 Introduction 2 Background and Related Work 3 Empirical Datasets 4 Experiment Setup 4.1 Research Questions 4.2 Experiment Process 4.3 Evaluation Metrics 5 Results and Discussion 6 Conclusions References Global Interpretable Calibration Index, a New Metric to Estimate Machine Learning Models' Calibration 1 Introduction 2 Background and Related Work 3 Definition of the GICI Index 4 GICI-Based Recalibration 5 Comparative Experiments 5.1 Datasets 5.2 Experimental Setup 5.3 Results and Implications 6 Conclusions References The ROC Diagonal is Not Layperson's Chance: A New Baseline Shows the Useful Area 1 Introduction 2 Binary Chance 3 The Binary Chance Baseline 4 The Main Diagonal 5 The Useful Area Under the ROC Curve 6 Prevalence and Costs May Counteract Each Other 7 Examples in Classification 8 Related Work 9 Conclusions and Future Work References Debiasing MDI Feature Importance and SHAP Values in Tree Ensembles 1 Variable Importance in Trees 2 Separating Inbag and Out-of-Bag (OOB) Samples 3 Conditional Feature Contributions (CFCs) 4 SHAP Values 4.1 Null/Power Simulations 4.2 Noisy Feature Identification 4.3 Shrunk SHAP 5 Discussion A Appendix A.1 Background and Notations A.2 Debiasing MDI via OOB Samples A.3 Variance Reduction View A.4 E(PG"0362PGoob(0)) = 0 References The Influence of User Diversity on Motives and Barriers when Using Health Apps - A Conjoint Investigation of the Intention-Behavior Gap 1 Introduction 2 Related Work 2.1 Relevance of Physical Activity and Health Apps 2.2 Understanding the Process of Health Behavior Change 2.3 Previous Findings on the Impact of Gender and Age on Health App Use 2.4 Scales Used in the Questionnaire 2.5 Underlying Research Questions 3 Method 3.1 Sample 3.2 The Questionnaire 3.3 Description of the Choice-Based Conjoint Study and Decision Scenarios 4 Results 4.1 Hierarchical Bayes Analysis 4.2 Latent Class Analysis 4.3 Decision-Making of the Groups 4.4 Gender, Age and Sport-Activity in the Groups 4.5 Further Influence of Gender, Age and Sport-Activity 4.6 Influence of Personality and Motivation Sources on Decision-Making 5 Discussion 5.1 Referring to the Health Action Process Approach 5.2 Limitations of the Study 6 Conclusion References Identifying Fraud Rings Using Domain Aware Weighted Community Detection 1 Introduction 2 Related Work 2.1 Community Detection 2.2 Graph-based Fraud Detection 3 Fraud Detection Problem 3.1 Identifying Fraud Rings 3.2 Incorporating Domain Knowledge 3.3 Other Challenges in Fraud Detection 4 Proposed Framework 4.1 Overview 4.2 Graph Construction 4.3 Community Detection 4.4 Downstream Task 5 Experiments 5.1 Experimental Setup 5.2 Experimental Settings 5.3 Implementation 5.4 Experimental Results 6 Conclusion and Future Work References Capabilities, Limitations and Challenges of Style Transfer with CycleGANs: A Study on Automatic Ring Design Generation 1 Introduction 2 Related Work 3 Proposed Model for Automatic Ring Design Generation 3.1 Practical Use Case: Designing XYU Rings: Traditional Ring Design Pipeline 3.2 Proposed Pipeline: Automatic Design Rendering Through Generative Models for Image-to-Image Translation 3.3 CycleGAN as a Generative Model Trained on Unpaired Images 4 Results and Analysis 4.1 Sketch2Rendering image results 5 Discussion 5.1 Challenging Aspects and Detected Artefacts 5.2 Model Limitations 6 Conclusions and Future Work A Appendix: Supplementary Materials A.1 Datasets A.2 CycleGAN Model A.3 CycleGAN Training details References Semantic Causal Abstraction for Event Prediction 1 Introduction 2 Related Work 3 Methods 3.1 Dataset 3.2 Causal Phrase Parsing 3.3 Semantic Representations 3.4 Semantic Causal Abstraction 3.5 Prediction Model 4 Experiments 4.1 Setup 4.2 Stock Trend Prediction 4.3 Volatility Prediction 4.4 Results 5 Conclusion References An Evaluation Study of Intrinsic Motivation Techniques Applied to Reinforcement Learning over Hard Exploration Environments 1 Introduction 2 Related Work and Contribution 3 Methodology of the Study 3.1 RQ1: Varying the Intrinsic Reward Coefficient 3.2 RQ2: Episodic State Counts Versus First-Visit Scaling 3.3 RQ3: Sensitiveness to the Neural Network Architectures 4 Experimental Setup 4.1 Environments 4.2 Baselines and Hyperparameters 4.3 Network Architectures 5 Results and Analysis 6 Conclusion References Towards Generating Financial Reports from Tabular Data Using Transformers 1 Introduction 2 Background 2.1 Financial Reports 2.2 Transformer Networks 3 Related and Previous Work 4 Implementation 5 Results 6 Challenges and Future Work 7 Conclusion References Evaluating the Performance of SOBEK Text Mining Keyword Extraction Algorithm 1 Introduction 2 SOBEK Text Mining 3 Methods 4 Results 5 Discussion and Final Considerations References Classification of Screenshot Image Captured in Online Meeting System 1 Introduction 2 Preliminaries 2.1 Encoding of Video Stream 2.2 Multimedia Forensics 2.3 Detection of Capturing Traces 3 Classification of Captured Images 3.1 CNN-based Binary Image Classifier 3.2 Training 3.3 Dataset 4 Experimental Results 4.1 Conditions 4.2 Evaluation Metrics 4.3 Classification 4.4 Classification After Anti-forensics 5 Concluding Remarks References A Survey on the Application of Virtual Reality in Event-Related Potential Research 1 Introduction 2 Research Methods and Inclusion Criteria 3 Findings 3.1 Integration of Virtual Reality and Electroencephalography 3.2 Experimental ERP Components and Paradigms 3.3 Technical Configurations 3.4 Virtual Reality for ERP Analysis and Application 3.5 Reliability of Findings 4 Final Remarks and Recommendations References Visualizing Large Collections of URLs Using the Hilbert Curve 1 Introduction 2 Related Work 2.1 Visualizing Textual Data 2.2 Visualization of Large Corpora 2.3 Monitoring of Online Media Streams 2.4 Map-Like Visualization 3 Method 3.1 Data: Requirements and Description 3.2 Problem 3.3 Why Not Map into a Metric Space? 3.4 Using Space-Filling Curves 3.5 Implementation 3.6 R Package 4 Demonstration 5 Discussion 6 Conclusion References How to Reduce the Time Necessary for Evaluation of Tree-Based Models 1 Introduction and Motivation 1.1 Related Works 2 Methods Improving the ML Models’ Interpretability 3 Evaluation Metrics of Model Interpretability and Quality 4 Proposed Approach 4.1 Use Case 4.2 Selected Metrics 4.3 Multiple-Criteria Decision Making 5 Conclusion and Future Work References An Empirical Analysis of Synthetic-Data-Based Anomaly Detection 1 Introduction 2 Related Work 3 Experiment Setting 3.1 Datasets 3.2 Dataset Synthetization 3.3 Dataset Pre-procesing 3.4 Anomaly Detection Methods 4 Results 4.1 Credit Card Dataset 4.2 Annthyroid Dataset 4.3 Observations 5 Conclusion A Appendix: Additional Results A.1 Credit Card Dataset Results A.2 Annthyroid Dataset Results References SECI Model in Data-Based Procedure for the Assessment of the Frailty State in Diabetic Patients 1 Introduction 1.1 SECI Model 1.2 Analytical Methods 1.3 Related Work 2 Results 3 Discussion 4 Conclusion References Comparing Machine Learning Correlations to Domain Experts' Causal Knowledge: Employee Turnover Use Case 1 Introduction 2 Predicting Employee Turnover Using Machine Leaning 2.1 Related Work 2.2 Testing Classification Algorithms on IBM HR Data 3 Causal Knowledge Relevance in Improving Employee Turnover Prediction 3.1 Causal Knowledge Discovery 3.2 Causal Links and Factors Importance Interpretation 4 Results Comparison 5 Discussion and Recommendation for Future Research 6 Conclusion References Machine Learning and Knowledge Extraction to Support Work Safety for Smart Forest Operations 1 Introduction 2 Dataset 2.1 Description 2.2 Preprocessing 3 Predictive Models 3.1 Regression Task 3.2 Classification 4 Conclusion and Future Research Questions References Author Index