Machine Learning for Civil and Environmental Engineers: A Practical Approach to Data-Driven Analysis, Explainability, and Causality

Description

Accessible and practical framework for machine learning applications and solutions for civil and environmental engineers This textbook introduces engineers and engineering students to the applications of artificial intelligence (AI), machine learning (ML), and machine intelligence (MI) in relation to civil and environmental engineering projects and problems, presenting state-of-the-art methodologies and techniques to develop and implement algorithms in the engineering domain. Through real-world projects like analysis and design of structural members, optimizing concrete mixtures for site applications, examining concrete cracking via computer vision, evaluating the response of bridges to hazards, and predicating water quality and energy expenditure in buildings, this textbook offers readers in-depth case studies with solved problems that are commonly faced by civil and environmental engineers. The approaches presented range from simplified to advanced methods, incorporating coding-based and coding-free techniques. Professional engineers and engineering students will find value in the step-by-step examples that are accompanied by sample databases and codes for readers to practice with. Written by a highly qualified professional with significant experience in the field, Machine Learning includes valuable information on: The current state of machine learning and causality in civil and environmental engineering as viewed through a scientometrics analysis, plus a historical perspectiveSupervised vs. unsupervised learning for regression, classification, and clustering problemsExplainable and causal methods for practical engineering problemsDatabase development, outlining how an engineer can effectively collect and verify appropriate data to be used in machine intelligence analysisA framework for machine learning adoption and application, covering key questions commonly faced by practitioners This textbook is a must-have reference for undergraduate/graduate students to learn concepts on the use of machine learning, for scientists/researchers to learn how to integrate machine learning into civil and environmental engineering, and for design/engineering professionals as a reference guide for undertaking MI design, simulation, and optimization for infrastructure. Machine Learning for Civil & Environmental Engineers Contents Preface About the Companion Website 1 Teaching Methods for This Textbook Synopsis 1.1 Education in Civil and Environmental Engineering 1.2 Machine Learning as an Educational Material 1.3 Possible Pathways for Course/Material Delivery 1.3.1 Undergraduate Students 1.3.2 Graduate Students and Post-docs 1.3.3 Engineers and Practitioners 1.3.4 A Note 1.4 Typical Outline for Possible Means of Delivery Chapter Blueprint Questions and Problems References 2 Introduction to Machine Learning Synopsis 2.1 A Brief History of Machine Learning 2.2 Types of Learning 2.3 A Look into ML from the Lens of Civil and Environmental Engineering 2.4 Let Us Talk a Bit More about ML 2.5 ML Pipeline 2.5.1 Formulating a Hypothesis 2.5.2 Database Development 2.5.3 Processing Observations 2.5.4 Model Development 2.5.5 Model Evaluation 2.5.6 Model Optimization 2.5.7 Model Deployment 2.5.8 Model Management (Monitoring, Updating, Etc.) 2.6 Conclusions Definitions Chapter Blueprint Questions and Problems References 3 Data and Statistics Synopsis 3.1 Data and Data Science 3.2 Types of Data 3.2.1 Numerical Data 3.2.2 Categorical Data 3.2.3 Footage 3.2.4 Time Series Data* 3.2.5 Text Data* 3.3 Dataset Development 3.4 Diagnosing and Handling Data 3.5 Visualizing Data 3.6 Exploring Data 3.6.1 Correlation-based and Information-based Methods 3.6.2 Feature Selection and Extraction Methods 3.6.3 Dimensionality Reduction 3.7 Manipulating Data 3.7.1 Manipulating Numerical Data 3.7.2 Manipulating Categorical Data 3.7.3 General Manipulation 3.8 Manipulation for Computer Vision 3.9 A Brief Review of Statistics 3.9.1 Statistical Concepts 3.9.2 Regression 3.10 Conclusions Definitions Chapter Blueprint Questions and Problems References 4 Machine Learning Algorithms Synopsis 4.1 An Overview of Algorithms 4.1.1 Supervised Learning 4.1.2 Unsupervised Learning 4.2 Conclusions Definitions Chapter Blueprint Questions and Problems References 5 Performance Fitness Indicators and Error Metrics Synopsis 5.1 Introduction 5.2 The Need for Metrics and Indicators 5.3 Regression Metrics and Indicators 5.4 Classification Metrics and Indicators 5.5 Clustering Metrics and Indicators 5.6 Functional Metrics and Indicators* 5.6.1 Energy-based Indicators 5.6.2 Domain-specific Metrics and Indicators 5.6.3 Other Functional Metrics and Indicators 5.7 Other Techniques (Beyond Metrics and Indicators) 5.7.1 Spot Analysis 5.7.2 Case-by-Case Examination 5.7.3 Drawing and Stacking 5.7.4 Rational Vetting* 5.7.5 Confidence Intervals* 5.8 Conclusions Definitions Chapter Blueprint Questions and Problems Suggested Metrics and Packages References 6 Coding-free and Coding-based Approaches to Machine Learning Synopsis 6.1 Coding-free Approach to ML 6.1.1 BigML 6.1.2 DataRobot 6.1.3 Dataiku 6.1.4 Exploratory 6.1.5 Clarifai 6.2 Coding-based Approach to ML 6.2.1 Python 6.2.2 R 6.3 Conclusions Definitions Chapter Blueprint Questions and Problems References 7 Explainability and Interpretability Synopsis 7.1 The Need for Explainability 7.1.1 Explainability and Interpretability 7.2 Explainability from a Philosophical Engineering Perspective* 7.3 Methods for Explainability and Interpretability 7.3.1 Supervised Machine Learning 7.3.2 Unsupervised Machine Learning 7.4 Examples 7.4.1 Surrogates* 7.4.2 Global Explainability 7.4.3 Local Explainability 7.5 Conclusions Definitions Questions and Problems Chapter Blueprint References 8 Causal Discovery and Causal Inference Synopsis 8.1 Big Ideas Behind This Chapter 8.2 Re-visiting Experiments 8.3 Re-visiting Statistics and ML 8.4 Causality 8.4.1 Definition and a Brief History 8.4.2 Correlation and Causation 8.4.3 The Causal Rungs 8.4.4 Regression and Causation 8.4.5 Causal Discovery and Causal Inference 8.4.6 Assumptions Required to Establish Causality 8.4.7 Causal Graphs and Graphical Methods 8.4.8 Causal Search Methods and ML Packages 8.4.9 Causal Inference and ML Packages 8.4.10 Causal Approach 8.5 Examples 8.5.1 Causal Discovery 8.5.2 Causal Inference 8.5.3 DAG from CausalNex 8.5.4 Modifying CausalNex’s DAG with Domain Knowledge 8.5.5 A DAG Similar to a Regression Model 8.6 A Note on Causality and ML 8.7 Conclusions Definitions Questions and Problems Chapter Blueprint References 9 Advanced Topics (Synthetic and Augmented Data, Green ML, Symbolic Regression, Mapping Functions, Ensembles, and AutoML) Synopsis 9.1 Synthetic and Augmented Data 9.1.1 Big Ideas 9.1.2 Conservative Interpolation 9.1.3 Synthetic Minority Over-sampling Technique (SMOTE) 9.1.4 Generative Adversarial Networks (GANs) and Triplet-based Variational Autoencoder (TVAE) 9.1.5 Augmented Data 9.1.6 A Note 9.2 Green ML 9.2.1 Big Ideas 9.2.2 Example 9.2.3 Energy Perspective 9.2.4 A Note 9.3 Symbolic Regression 9.3.1 Big Ideas 9.3.2 Examples 9.3.3 Eureqa 9.3.4 TurningBot 9.3.5 HeuristicLab 9.3.6 GeneXproTools⁎ 9.3.7 Online Interface by MetaDemoLab 9.3.8 Python 9.3.9 Eureqa 9.3.10 MetaDemoLab 9.3.11 Python 9.3.12 GeneXproTools* 9.3.13 Eureqa 9.3.14 MetaDemoLab 9.3.15 HeuristicLab 9.3.16 A Note 9.4 Mapping Functions 9.4.1 Big Ideas 9.4.2 Concept of Mapping Functions 9.4.3 Approach to Mapping Functions 9.4.4 Example 9.4.5 A Note 9.5 Ensembles 9.5.1 Big Ideas 9.5.2 Examples 9.6 AutoML 9.6.1 Big Ideas 9.6.2 The Rationale and Anatomy of CLEMSON 9.6.3 Example 9.6.4 A Note 9.7 Conclusions Definitions Questions and Problems Chapter Blueprint References 10 Recommendations, Suggestions, and Best Practices Synopsis 10.1 Recommendations 10.1.1 Continue to Learn 10.1.2 Understand the Difference between Statistics and ML 10.1.3 Know the Difference between Prediction via ML and Carrying Out Tests and Numerical Simulations 10.1.4 Ask if You Need ML to Address the Phenomenon on Hand 10.1.5 Establish a Crystal Clear Understanding of Model Assumptions, Outcomes, and Limitations 10.1.6 Remember that an Explainable Model Is Not a Causal Model 10.1.7 Master Performance Metrics and Avoid the Perception of False Goodness 10.1.8 Acknowledge that Your Model Is Likely to Be Biased 10.1.9 Consult with Experts and Re-visit Domain Knowledge to Identify Suitable Features 10.1.10 Carefully Navigate the Trade-offs 10.1.11 Share Your Data and Codes 10.2 Suggestions 10.2.1 Start Your Analysis with Simple Algorithms 10.2.2 Explore Algorithms and Metrics 10.2.3 Be Conscious of Data Origin 10.2.4 Emphasize Model Testing 10.2.5 Think Beyond Training and Validation 10.2.6 Trace Your Model Beyond Deployment 10.2.7 Convert Your ML Models into Web and Downloadable Applications 10.2.8 Whenever Possible, Include Physics Principles in ML Models 10.3 Best Practices 10.3.1 Avoid the Use of “Small” and Low Quality Data 10.3.2 Be Aware of the Most Commonly Favored ML Algorithms 10.3.3 Follow the Most Favored Model Development Procedures 10.3.4 Report Statistics on Your Dataset 10.3.5 Avoid Blackbox Models in Favor of Explainable and Causal Models (Unless the Goal Is to Create a Blackbox Model) 10.3.6 Integrate ML into Your Future Works Definitions Questions and Problems References 11 Final Thoughts and Future Directions Synopsis 11.1 Now 11.2 Tomorrow 11.2.1 Big, Small, and Imbalanced Data 11.2.2 Learning ML 11.2.3 Benchmarking ML 11.2.4 Standardizing ML 11.2.5 Unboxing ML 11.2.6 Popularizing ML 11.2.7 Engineering ML 11.3 Possible Ideas to Tackle 11.4 Conclusions References Index EULA