Indiana University-Purdue University Indianapolis (IUPUI) / After the initial network is constructed using expert’s knowledge of the domain, Bayesian networks need to be updated as and when new data is observed. Literature mining is a very important source of this new data. In this work, we explore what kind of data needs to be extracted with the view to update Bayesian Networks, existing technologies which can be useful in achieving some of the goals and what research is required to accomplish the remaining requirements. This thesis specifically deals with utilizing causal associations and experimental results which can be obtained from literature mining. However, these associations and numerical results cannot be directly integrated with the Bayesian network. The source of the literature and the perceived quality of research needs to be factored into the process of integration, just like a human, reading the literature, would. This thesis presents a general methodology for updating a Bayesian Network with the mined data. This methodology consists of solutions to some of the issues surrounding the task of integrating the causal associations with the Bayesian Network and demonstrates the idea with a semiautomated software system.

Advances in information technology have presented the potential to automate investment decision making processes. This will alleviate the need for manual analysis and reduce the subjective nature of investment decision making. However, there are different investment approaches and perspectives for investing which makes acquiring and representing expert knowledge for share evaluation challenging. Current decision models often do not reflect the real investment decision making process used by the broader investment community or may not be well-grounded in established investment theory. This research investigates the efficacy of using ontologies and Bayesian networks for automating share evaluation on the JSE. The knowledge acquired from an analysis of the investment domain and the decision-making process for a value investing approach was represented in an ontology. A Bayesian network was constructed based on the concepts outlined in the ontology for automatic share evaluation. The Bayesian network allows decision makers to predict future share performance and provides an investment recommendation for a specific share. The decision model was designed, refined and evaluated through an analysis of the literature on value investing theory and consultation with expert investment professionals. The performance of the decision model was validated through back testing and measured using return and risk-adjusted return measures. The model was found to provide superior returns and risk-adjusted returns for the evaluation period from 2012 to 2018 when compared to selected benchmark indices of the JSE. The result is a concrete share evaluation model grounded in investing theory and validated by investment experts that may be employed, with small modifications, in the field of value investing to identify shares with a higher probability of positive risk-adjusted returns.

Due to the growing number of diverse power systems disruptions, including extreme weather events, technical factors, and human factors, assessing and quantifying the resilience of electric power subsystems has become an indispensable step to develop an efficient strategic plan to enhance the resilience and reliability of these systems and to endure the diverse interruptions. In this study, factors and sub-factors that may have either direct or indirect impact on the resilience of biomass-based combined heat and power systems are identified, and the interdependencies among them are determined as well. A Bayesian network model is implemented to quantify the resilience of a bCHP system, and the results are analyzed by applying three different techniques, which are sensitivity analysis, forward propagation analysis, and backward propagation analysis.

Ruminant livestock play a vital role in fulfilling the nutrient requirements of humans by providing protein, energy, and essential microminerals. With the increasing demand for meat and dairy products, the ruminant industry must continue to improve the productivity and efficiency of ruminant animals with limited resources while minimizing the environmental impact. Rumen fermentation is the focal point of the productivity and efficiency of the animal and numerous chemical, physical and biochemical interactions make the rumen a complex ecosystem. Therefore, improving the understanding of fermentation dynamics in a holistic manner and characterizing how fermentation varies in response to different nutrient supplies can greatly expand our knowledge on rumen fermentation to develop better engineered rumen manipulation strategies. The central aim of these investigations was to employ alternative analytical strategies for holistic exploration of complex relationships among rumen, animal, and dietary variables and to estimate rumen volatile fatty acid (VFA) dynamics under different nutrient supplies. The objective of the first study was to explore the strengths and limitations of mixed-model meta-analysis, recursive feature elimination (RFE), and additive Bayesian networking (ABN) in identifying relationships among diet, rumen, and milk performance variables. Both mixed-models and ABN agreed upon most of the variables and relationships identified while RFE failed to capture interactions. Given the capacity of mixed models for quantitative inquiry and the potential of ABN to illustrate complex associations in a more intuitive way, future investigations combining both approaches hold potential to explore intercorrelated data in a holistic manner. Followed by the successful use of ABN in the first study, the goal of a follow up study was to investigate the potential of two different network approaches to explore rumen level interactions using data generated in continuous culture experiments. Two network analysis approaches, EBIC-LASSO network (ELN) and Bayesian learning network (BLN) were leveraged to explore the relationships among rumen fermentation parameters in continuous culture experiments. Unidirectional ELN illustrated prominent variables while BLN, which produces a directed acyclic graph, identified directional relationships implying causality. Overall, both networking approaches demonstrate strengths in capturing connectedness and directionality of rumen fermentation variables. In a complementary line of work, the next experiment focused on developing an alternative method for iso-tope based assessments to produce less expensive, and more efficient screening of fermentation conditions driven by diet. Cannulated wethers were used in this study and 4 dietary treatments combining lowly and highly degradable fiber (timothy hay and beet pulp, respectively) and protein (heat-treated soybean meal and soybean meal, respectively) were tested. Results indicated that fluid volume of the rumen and the rate of passage were influenced by protein, but not fiber, source. Higher rumen volumes and lower passage rates were associated with heat-treated soybean meals. The effect of dietary treatments on VFA absorption dynamics was prominent compared to the minimal changes in production dynamics. Overall, heat-treated soybean meal appears to influence VFA disappearance resulting in low concentrations within the rumen, but greater flux of VFA disappearance. In conclusion, this method demonstrated the capacity to estimate VFA dynamics beyond concentrations and molar proportions while being cost effective and more physiologically relevant. In a fourth study, we sought to investigate the growth performance and rumen VFA profile in response to different planes of nutrients and naturally occurring coccidiosis. Coccidiosis infection altered rumen isobutyrate concentrations and tended to alter major VFA concentrations suggesting the need of future work to explore coccidiosis effects on rumen fermentation. The first two investigations highlighted the potential and strength of leveraging alternative analytical tools to complement statistical approaches generally used in ruminant nutrition while concurrently improving ability to explain complex associations in the rumen. The third and fourth projects characterized the rumen VFA dynamics and profile in response to the different nutrient degradability and health status, respectively. Collectively, these investigations contribute to better understanding of rumen dynamics through novel analytical and experimental approaches. / Doctor of Philosophy / With increasing global population, income, urbanization, and changes in dietary habits, the demand for meat and milk continues to grow. The ruminant animal industries (beef cattle, dairy cattle, sheep, goat, and buffalo) carry the burden of increasing production utilizing limited resources while minimizing the negative environmental impact caused by ruminant operations. To achieve this goal the productivity of the animal must be increased, and in order to increase the efficiency of production a better understanding of factors driving the production is critical. Ruminant animals have the unique ability to convert plant fiber into human edible milk and meat through a process that predominantly occurs in the special gut compartment called the rumen. In this process several compounds are produced, and among those volatile fatty acid (VFA) is of utmost importance because it fulfills energy demands for growth, production and reproduction. The rumen is a complex ecosystem consisting of numerous variables and associations. Understanding those relationships is crucial to manipulate rumen mechanisms. The overall objective of this work was to evaluate the potential of alternative statistical approaches, which demonstrated success in other disciplines, for better depiction of complex associations and characterization of production and absorption mechanisms of rumen VFA in response to different nutrient supplies. The objective of first investigation was to evaluate a feature selection method (recursive feature elimination; RFE) and a network approach (Additive Bayesian network; ABN) concurrently with a standard variable selection method (mixed model meta-analysis) commonly used to develop animal nutrition models. We attempted to find out the most important dietary, rumen, and animal variables for milk yield, milk fat and protein content as an example. Results indicate that the network approach was well aligned with the standard tool and can be used as a complementary approach. In our second investigation, we leveraged two networking analyses, a frequentist network which was unidirectional and a Bayesian network which was directional to explore rumen level interactions. The unidirectional network approach highlighted the most important variables in the rumen and numerous relationships among these variables. The directional network was more useful in understanding of causal associations within the system. In the third experiment we estimated the production and absorption of VFA in response to the different protein (heat-treated and regular soybean meal) and fiber (timothy hay and beet pulp) sources. The results revealed that the production of VFA was minimally affected by the diet, but the absorption was higher with heat-treated soybean meal. Our last project investigated the effect of parasitic infection, i.e., coccidiosis, and high and low levels of nutrition on growth and rumen VFA of growing lambs. Infection of coccidiosis altered a minor VFA (isobutyrate) and tended to alter total and major VFA (acetate and propionate). All these findings help to improve our understanding of rumen fermentation and subsequently develop strategies to manipulate rumen fermentation to enhance efficiency and productivity.

The Precast/Prestressed Concrete Institute (PCI) is the technical institute for precast/prestressed concrete industry. The plant involves activities such as placing high tensile steel strings inside the concrete products before they harden. This process needs the strings to be "stressed" hydraulically with high tension, which provides possibility of breaking the strand. Hence, employees may face a severe injury around the stressing bed. As various activities take place on the plant at the same time, employees must follow certain safety protocols while being around the plant. Another safety concern on the precast plant is silica exposure. Occupational Safety and Health Administration (OSHA) has provided various guidelines and tools to minimize silica exposure. Employees need to be careful and follow these safety protocols, otherwise it may lead to severe lung disease. Thus, employees need the appropriate safety training which will motivate them to follow safety protocols rigorously. The Bayesian Network (BN) methodology helps analyze plant structure to understand potential risk factors and causes that can be fixed by the employer paying more attention. The current traditional training methods such as videos, PowerPoint slides, or on-paper training, are not as effective in conveying the severity of the risky situations. This research focuses on precast plant activities while trying to identify the factors affecting plant safety. The current results suggest that using the BN study for the factors, such as stressing, chipping, leg injuries, tripping, and suspended loads, that may cause accidents or affect plant safety have a major impact on overall plant safety. Further sections of the dissertation discuss Fault Tree Analysis for risk assessment. It is observed that the BN study outperforms the risk assessment. Improvisation in safety protocols associated with these factors will help mitigate overall plant risks. In addition, study includes the development of immersive training methods and comparison of the immersive method to current safety training methods. Virtual Reality (VR) training module provides significant evidence to improvement in motivation level compared to traditional training. Knowledge gain concerning the safety protocols proves to be increasing for employees after the VR training method compared to the traditional training methods.

In this dissertation, we describe a new approach for gene finding that can utilize proteomics information in addition to DNA and RNA to identify new genes in prokaryote genomes. Proteomics processing pipelines require identification of small pieces of proteins called peptides. Peptide identification is a very error-prone process and we have developed a new algorithm for validating peptide identifications using a distance-based outlier detection method. We demonstrate that our method identifies more peptides than other popular methods using standard mixtures of known proteins. In addition, our algorithm provides a much more accurate estimate of the false discovery rate than other methods. Once peptides have been identified and validated, we use a second algorithm, proteogenomic mapping (PGM) to map these peptides to the genome to find the genetic signals that allow us to identify potential novel protein coding genes called expressed Protein Sequence Tags (ePSTs). We then collect and combine evidence for ePSTs we generated, and evaluate the likelihood that each ePST represents a true new protein coding gene using supervised machine learning techniques. We use machine learning approaches to evaluate the likelihood that the ePSTs represent new genes. Finally, we have developed new approaches to Bayesian learning that allow us to model the knowledge domain from sparse biological datasets. We have developed two new bootstrap approaches that utilize resampling to build networks with the most robust features that reoccur in many networks. These bootstrap methods yield improved prediction accuracy. We have also developed an unsupervised Bayesian network structure learning method that can be used when training data is not available or when labels may not be reliable.

Bayesian network is a popular machine learning tool for modeling uncertain dependence relationships among the random factors of a domain. It represents the relations qualitatively by using a directed acyclic graph (DAG) and quantitatively by using a set of conditional probability distributions. Several exact algorithms for learning optimal Bayesian networks from data have been developed recently. However, these algorithms are still inefficient to some extent. This is not surprising because learning Bayesian network has been proven to be an NP-Hard problem. Based on a critique of these algorithms, this thesis introduces a new algorithm based on heuristic search for learning optimal Bayesian.

No description available.

No description available.

No description available.