Bayesian Model Averaging (BMA) is a new area in modern applied statistics that provides data analysts with an efficient tool for discovering promising models and obtaining esti-mates of their posterior probabilities via Markov chain Monte Carlo (MCMC). These probabilities can be further used as weights for model averaged predictions and estimates of the parameters of interest. As a result, variance components due to model selection are estimated and accounted for, contrary to the practice of conventional data analysis (such as, for example, stepwise model selection). In addition, variable activation probabilities can be obtained for each variable of interest. This dissertation is aimed at connecting BMA and various ramifications of the multivari-ate technique called Reduced-Rank Regression (RRR). In particular, we are concerned with Canonical Correspondence Analysis (CCA) in ecological applications where the data are represented by a site by species abundance matrix with site-specific covariates. Our goal is to incorporate the multivariate techniques, such as Redundancy Analysis and Ca-nonical Correspondence Analysis into the general machinery of BMA, taking into account such complicating phenomena as outliers and clustering of observations within a single data-analysis strategy. Traditional implementations of model averaging are concerned with selection of variables. We extend the methodology of BMA to selection of subgroups of observations and im-plement several approaches to cluster and outlier analysis in the context of the multivari-ate regression model. The proposed algorithm of cluster analysis can accommodate re-strictions on the resulting partition of observations when some of them form sub-clusters that have to be preserved when larger clusters are formed. / Ph. D.

京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第21080号 / 工博第4444号 / 新制||工||1691(附属図書館) / 京都大学大学院工学研究科社会基盤工学専攻 / (主査)教授 KIM Chul-Woo, 教授 杉浦 邦征, 教授 八木 知己 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

In today's interconnected infrastructures, Cyber-Physical Systems (CPSs) play a critical role in domains including water distribution, agricultural production, and energy management. Modern infrastructures rely on a network of cyber-physical components—mechanical actuators, electrical sensors, and internet-connected devices—to supervise and manage operational processes. However, the increasing complexity and connectivity of these systems amplify their vulnerability to cyberattacks, necessitating robust cybersecurity measures and effective Outlier Detection (OD) methods. These methods are essential to prevent infrastructure failures, reduce environmental waste, and mitigate damages caused by malicious activities. Existing approaches often lack the integration of multiple operational metrics and context-driven techniques, hampering their effectiveness in real-world scenarios. In large CPSs—comprising hundreds or thousands of sensors, actuators, PLCs, IoT devices, and complex Control and Protection Switching Gear (CPSG)—the challenge of ensuring data quality, security, and reliability is costly. Cyberattacks frequently appear as outliers or anomalies in the data and are launched with "minimum perturbation," making their detection significantly challenging. This dissertation proposes a novel framework, multiple pipelines, and AI-based methods to develop context-driven, data-driven, and assurance-focused OD solutions. Emphasis is placed on water and agricultural systems, illustrating the proposed framework's effectiveness, particularly through enhanced decision-making, operational efficiency, and cybersecurity measures. A comprehensive survey of OD methods that employ Artificial Intelligence (AI) techniques establishes the foundational understanding of OD. This survey underscores that successful OD depends on domain knowledge, contextual factors, and assurance principles. Synthesizing these insights, the dissertation leverages synthetically generated SCADA data and GAN-produced poisoned data, as well as real-world SCADA data from Wastewater Treatment Plants (WWTPs), to identify outliers and address critical problems—such as forecasting tunnel wastewater overflows under extreme weather conditions—by applying Recurrent Neural Network (RNN)-based Deep Learning (DL) methods. Additionally, an AI-based decision support tool is introduced to detect anomalies in complex plant data and optimize operational set-points, thereby aiding Operation and Maintenance (OandM) in Water Distribution Systems (WDSs). Similarly, in Agricultural Production Systems (APSs), which traditionally rely on reactive policies and short-term solutions, integrating advanced AI-driven OD methods provides farmers with timely, data-informed decisions that account for contextual changes resulting from outlier events. Machine Learning (ML) and DL methods measure associations, correlations, and causations among global and domestic factors, aiding in the accurate prediction of agricultural production. This contextual awareness helps manage policy, optimize resource utilization, and support precision agriculture strategies. The main contributions of this dissertation include introducing a novel framework that integrates OD techniques with AI assurance and context-driven methodologies in CPSs; developing multiple pipelines and DL models that enhance anomaly detection, forecasting accuracy, and proactive decision support in WDSs and APSs; and demonstrating measurable improvements in cybersecurity, operational efficiency, and predictive capability using real-world and synthetic data. These efforts collectively foster more trustworthy and sustainable CPSs. Experimental results are recorded, evaluated, and discussed, revealing that these contributions bridge the gap between complex theoretical constructs and tangible real-world applications. / Doctor of Philosophy / Recent unprecedented AI and sensor technology advancements are transforming all domains, including Water Distribution Systems (WDSs) and Agricultural Production Systems (APSs). With Industry 4.0, WDSs and APSs are undergoing a significant digital transformation to enable data-driven monitoring and control of utility operations. Incorporating cyber elements—such as sensors, actuators, data transmitters, receivers, Programmable Logic Controllers (PLCs), and Internet of Things (IoT) devices—aims to make these Cyber-Physical Systems (CPSs) more effective in Operation and Maintenance (OandM). However, this progress comes with a trade-off, as CPSs become increasingly vulnerable to security and safety threats. For example, in 2013, hackers seized control of a small Florida dam, releasing unprocessed water into nearby communities. Furthermore, on February 5th, 2021, a Florida water treatment plant (in Oldsmar, FL) was compromised when the hacker altered the levels of sodium hydroxide (NaOH) in the water—a chemical that would severely damage human tissue. Recent targeted attacks on infrastructure in Ukraine also highlight the risks facing critical infrastructures worldwide, including WDSs. These events suggest that current control operations are largely exposed, necessitating sophisticated learning algorithms that can estimate system states, detect anomalies, and mitigate the harm caused by such intrusions. Technology has fundamentally transformed agriculture as well, significantly impacting this domain. Agriculture, a vital occupation in numerous countries, now faces increasing global population pressures. The United Nations (UN) projects the population to reach 9.7 billion by 2050, intensifying the strain on limited arable land. With only a 4% increase in cultivable land expected by 2050, farmers must do more with less. Traditional methods are insufficient to meet the soaring demands, as a 60% increase in food production is needed to feed an additional two billion people. This necessity for enhanced productivity and reduced waste drives the integration of AI into the agricultural sector. AI adoption not only accelerates efficiency but also increases production volumes, shortening the time from farm to market. This dissertation proposes novel, data- and context-driven Deep Learning (DL)-based methods and decision-support tools to enhance cybersecurity and anomaly detection within WDSs and APSs. Focusing on these critical infrastructures demonstrates how AI-driven strategies can effectively address real-world challenges and improve resilience, operational efficiency, and overall trustworthiness. The contributions of this dissertation include a framework and pipelines that incorporate contextual insights and AI assurance principles to improve anomaly detection and cybersecurity in these domains; the development of DL models tailored for identifying complex outliers and providing actionable decision-support, thereby optimizing resource allocation and ensuring sustainable operations; and validation of these approaches through experimental evaluations using real-world and synthetic data. Collectively, these efforts highlight significant improvements in reliability, efficiency, and scalability for critical infrastructure management, bridging the gap between theoretical advances in AI-driven anomaly detection and their practical application in WDSs and APSs.

This research explores the potential of improving the explainability of outliers using Bayesian Belief Networks as background knowledge. Outliers are deviations from the usual trends of data. Mining outliers may help discover potential anomalies and fraudulent activities. Meaningful outliers can be retrieved and analyzed by using domain knowledge. Domain knowledge (or background knowledge) is represented using probabilistic graphical models such as Bayesian belief networks. Bayesian networks are graph-based representation used to model and encode mutual relationships between entities. Due to their probabilistic graphical nature, Belief Networks are an ideal way to capture the sensitivity, causal inference, uncertainty and background knowledge in real world data sets. Bayesian Networks effectively present the causal relationships between different entities (nodes) using conditional probability. This probabilistic relationship shows the degree of belief between entities. A quantitative measure which computes changes in this degree of belief acts as a sensitivity measure . The first contribution of this research is enhancing the performance for measurement of sensitivity based on earlier research work, the Interestingness Filtering Engine Miner algorithm. The algorithm developed (IBOX - Interestingness based Bayesian outlier eXplainer) provides progressive improvement in the performance and sensitivity scoring of earlier works. Earlier approaches compute sensitivity by measuring divergence among conditional probability of training and test data, while using only couple of probabilistic interestingness measures such as Mutual information and Support to calculate belief sensitivity. With ingrained support from the literature as well as quantitative evidence, IBOX provides a framework to use multiple interestingness measures resulting in better performance and improved sensitivity analysis. The results provide improved performance, and therefore explainability of rare class entities. This research quantitatively validated probabilistic interestingness measures as an effective sensitivity analysis technique in rare class mining. This results in a novel, original, and progressive research contribution to the areas of probabilistic graphical models and outlier analysis.

Clinical observations of individual patients are the cornerstones for furthering our understanding of the human body, diseases, and therapeutics. Traditionally, clinical observations were communicated through publishing case reports and case series. The effort of identifying and investigating unusual clinical observations has always rested on the shoulders of busy clinicians. To date, there has been little effort dedicated to increasing the efficiency of identifying unique and uncommon patient observations that may lead to valuable discoveries. In this thesis, we propose and implement an augmented intelligence framework to identify potential novel clinical observations by combining machine analytics through outlier analysis with the judgment of subject-matter experts. Preeclampsia is a significant cause of maternal and perinatal mortality and morbidity, and advances in its management have been slow. Considering the complex etiological nature of preeclampsia, clinical observations are essential in advancing our understanding of the disease and therapeutic approaches. Thus, the objectives and studies in this thesis aim to answer the hypothesis that using outlier analysis in preeclampsia-related medical data would lead to identifying previously uninvestigated clinical cases with new clinical insight. This thesis combines three articles published or submitted for publication in peer-reviewed journals. The first article (published) is a systematic review examining the extent to which case reports and case series in preeclampsia have contributed new knowledge or discoveries. We report that under one-third of the identified case reports and case series presented new knowledge. In our second article (submitted for publication), we provide an overview of outlier analysis and introduce the framework of augmented intelligence using our proposed extreme misclassification contextual outlier analysis approach. Furthermore, we conduct a systematic review of obstetrics-related research that used outlier analysis to answer scientific questions. Our systematic review findings indicate that such use is in its infancy. In our third article (published), we implement the proposed augmented intelligence framework using two different outlier analysis methods on two independent datasets from separate studies in preeclampsia and hypertensive disorders of pregnancy. We identify several clinical observations as potential novelties, thus supporting the feasibility and applicability of outlier analysis to accelerate clinical discovery.

No description available.