Probabilistic Bayesian Network Research Articles

Statistical modelling of networked human-automation performance using working memory capacity

This study examines the challenging problem of modelling the interaction between individual attentional limitations and decision-making performance in networked human-automation system tasks. Analysis of real experimental data from a task involving networked supervision of multiple unmanned aerial vehicles by human participants shows that both task load and network message quality affect performance, but that these effects are modulated by individual differences in working memory (WM) capacity. These insights were used to assess three statistical approaches for modelling and making predictions with real experimental networked supervisory performance data: classical linear regression, non-parametric Gaussian processes and probabilistic Bayesian networks. It is shown that each of these approaches can help designers of networked human-automated systems cope with various uncertainties in order to accommodate future users by linking expected operating conditions and performance from real experimental data to observable cognitive traits like WM capacity.Practitioner Summary: Working memory (WM) capacity helps account for inter-individual variability in operator performance in networked unmanned aerial vehicle supervisory tasks. This is useful for reliable performance prediction near experimental conditions via linear models; robust statistical prediction beyond experimental conditions via Gaussian process models and probabilistic inference about unknown task conditions/WM capacities via Bayesian network models.

A psychological approach to learning causal networks

We examine the role of a common cognitive heuristic in unsupervised learning of Bayesian probability networks from data. Human beings perceive a larger association between causal than diagnostic relationships. This psychological principal can be used to orient the arcs within Bayesian networks by prohibiting the direction that is less predictive. The heuristic increased predictive accuracy by an average of 0.51 % percent, a small amount. It also increased total agreement between different network learning algorithms (Max Spanning Tree, Taboo, EQ, SopLeq, and Taboo Order) by 25 %. Prior to use of the heuristic, the multiple raters Kappa between the algorithms was 0.60 (95 % confidence interval, CI, from 0.53 to 0.67) indicating moderate agreement among the networks learned through different algorithms. After the use of the heuristic, the multiple raters Kappa was 0.85 (95 % CI from 0.78 to 0.92). There was a statistically significant increase in agreement between the five algorithms (alpha < 0.05). These data suggest that the heuristic increased agreement between networks learned through use of different algorithms, without loss of predictive accuracy. Additional research is needed to see if findings persist in other data sets and to explain why a heuristic used by humans could improve construct validity of mathematical algorithms.

Prediction of road accidents: comparison of two Bayesian methods

In this paper, two Bayesian methods for the development of accident prediction models are compared: the well-acknowledged Empirical Bayes (EB) method and a recently developed method based on Bayesian Probabilistic Networks (BPNs). Brief descriptions of the two methods are provided and their commonalities, differences, advantages and disadvantages are discussed. Both methods can be used to develop models for the multivariate prediction of accident events and can be included in road infrastructure safety management systems such as road safety impact assessments or road safety audits. Using a comprehensive data-set taken from the Austrian rural motorway network, it is shown that the predictions of both models are in good accordance with the data. It is observed that the BPN models show a higher degree of correlation with the data than the models developed using the EB method, as measured through the higher values of the correlation coefficients (∼5–10%).

Effectiveness and Subjectivity of Visual Inspection as a Method to Assess Bending Stiffness and Strength of Chestnut Elements

Besides the difficulty of assessing an existing timber structure on site, the efficiency and accuracy of visual inspection is often compromised by its subjective nature inherited by the level of expertise of the inspector. This often leads to conservative predictions of the mechanical properties, even with the use of specific visual grading norms. The main objectives of this work are to assess the effectiveness of visual inspection as a method to define different classes of strength and stiffness and to provide a statistical analysis on its subjectivity. For that aim, visual inspection using Italian standard UNI 11119:2004 and bending tests of 20 old chestnut beams (Castanea sativa Mill.) at different scale element, were carried out. Comparisons and effectiveness of visual inspection is analyzed within and between different scales of the timber members, and also regarding different level of expertise of inspectors. The results evidence similar percentages of segments classified with higher and lower visual inspections classes and proved to be a good qualitative indicator of bending strength between sawn beams. An overall 42% accuracy of the most experienced inspectors was found with better differentiation between visual classes, whereas lower level inspectors scored approximately less 5%. Lower level inspectors also evidenced higher concentration of values around a higher mean for each class denoting a more conservative approach. Regardless of the inspector level, knot size was considered the main limiting visual parameter with higher influence in small scales of the timber elements. When studying the characterization of a single knot, coefficients of variation of 15.7% and 21.8% were found for measuring the minimum and maximum diameter. Bayesian probability networks were considered as to individually assess the accuracy in stiffness prediction of different level of inspectors, and by combination of their information, evidencing that parallel combination for prior information may allow the increase in visual inspection accuracy.

Complexity of Inferences in Polytree-shaped Semi-Qualitative Probabilistic Networks

Semi-qualitative probabilistic networks (SQPNs) merge two important graphical model formalisms: Bayesian networks and qualitative probabilistic networks. They provide a very general modeling framework by allowing the combination of numeric and qualitative assessments over a discrete domain, and can be compactly encoded by exploiting the same factorization of joint probability distributions that are behind the Bayesian networks. This paper explores the computational complexity of semi-qualitative probabilistic networks, and takes the polytree-shaped networks as its main target. We show that the inference problem is coNP-Complete for binary polytrees with multiple observed nodes. We also show that inferences can be performed in time linear in the number of nodes if there is a single observed node. Because our proof is constructive, we obtain an efficient linear time algorithm for SQPNs under such assumptions. To the best of our knowledge, this is the first exact polynomial-time algorithm for SQPNs. Together these results provide a clear picture of the inferential complexity in polytree-shaped SQPNs.

On Budgeted Influence Maximization in Social Networks

Given a fixed budget and an arbitrary cost for selecting each node, the budgeted influence maximization (BIM) problem concerns selecting a set of seed nodes to disseminate some information that maximizes the total number of nodes influenced (termed as influence spread) in social networks at a total cost no more than the budget. Our proposed seed selection algorithm for the BIM problem guarantees an approximation ratio of (1-1/√e). The seed selection algorithm needs to calculate the influence spread of candidate seed sets, which is known to be #P-complex. Identifying the linkage between the computation of marginal probabilities in Bayesian networks and the influence spread, we devise efficient heuristic algorithms for the latter problem. Experiments using both large-scale social networks and synthetically generated networks demonstrate superior performance of the proposed algorithm with moderate computation costs. Moreover, synthetic datasets allow us to vary the network parameters and gain important insights on the impact of graph structures on the performance of different algorithms.

Bottom damage scenarios for the hull girder structural assessment

This article covers the reliability assessment of the hull girder of a crude oil tanker, referring to a scenario in which the ship is exposed to sea loads after a damage to the bottom of the hull has occurred. A number of possible flooding configurations are examined, each one caused by a group of damage cases, characterized by different location and extent. Static loads, wave loads and residual structural resistance are determined for each damage case, with the objective of obtaining a prediction for the probability of the hull girder's failure. The various damage cases are compared to each other and unconditioned to derive the probability of failure extended to the ship's life due to a generic bottom damage.A probabilistic Bayesian Network model has been created to deal with these variables and with the dependency relationships existing between them. The results provided by the model are analyzed with the aim of identifying the parameters most influencing the problem. The work is intended to contribute to the development of a more rational treatment of accidental conditions in design structural requirements for ships.

Risk assessment for structural design criteria of FPSO systems. Part II: Consequence models and applications to determination of target reliabilities

The paper presents applications of the framework for generic risk assessment for FPSO systems described in a companion paper [1]. Consequence models for structural risk assessment are developed for failure of hull girder sections, hull components and mooring lines using Bayesian Probabilistic Networks (BPN). The estimation of conditional probabilities required in the BPN models, such as the probability of a combined failure of several components and system failure conditional on components failure, are discussed. The risk assessment framework is applied for decision making on target reliabilities for ULS design of mooring lines. The target reliabilities are found by optimizing expected life-cycle costs and verifying compliance with risk acceptance criteria developed according to the Life Quality Index (LQI) principle. It is shown that the optimal reliability index, which minimizes total expected costs, may not always lead to an acceptable solution from the point-of-view of the LQI criterion.

Risk assessment for structural design criteria of FPSO systems. Part I: Generic models and acceptance criteria

A generic framework for consequence assessment and risk analysis of FPSO systems for the purpose of establishing structural design criteria is introduced, taking basis in recent work by the Joint Committee on Structural Safety (JCSS) addressing the issue of system representation through exposure events, direct and indirect failure consequences. The scenarios considered for risk-based calibration of a design code safety format for FPSO systems are outlined. It is shown how these scenarios may be represented in a generic risk assessment model using Bayesian Probabilistic Networks (BPNs). Risk acceptance criteria related to consequences to humans are determined based on the Life Quality Index (LQI), which is outlined and discussed in this paper. The generic risk framework and acceptance criteria are then applied in a companion paper [1] to build consequence models and to determine target reliability indices for structural design of FPSO components.

Bayesian probabilistic network modeling from multiple independent replicates

Often protein (or gene) time-course data are collected for multiple replicates. Each replicate generally has sparse data with the number of time points being less than the number of proteins. Usually each replicate is modeled separately. However, here all the information in each of the replicates is used to make a composite inference about signal networks. The composite inference comes from combining well structured Bayesian probabilistic modeling with a multi-faceted Markov Chain Monte Carlo algorithm. Based on simulations which investigate many different types of network interactions and experimental variabilities, the composite examination uncovers many important relationships within the networks. In particular, when the edge's partial correlation between two proteins is at least moderate, then the composite's posterior probability is large.

Driver’s Intention Estimation Based on Bayesian Networks for a Highly-Safe Intelligent Vehicle

Highly safe intelligent vehicles can significantly reduce vehicle accidents by warning drivers of dangerous situations. Trajectory estimation of target vehicles is expected to be used in highly safe intelligent vehicles. Trajectory estimation requires that we estimate driver intent not detectable by sensors. The Bayesian Network (BN) building we propose for trajectory estimation related to driver intent defines driver intent hierarchically to simplify the BN as much as possible. Causal driver-intent relationships are discussed reflecting real-world motion. This raises the quality of driver-intent estimation and increasing inference performance. Experimental learning based on 2D image processing is presented to acquire probabilistic BN parameters.

Risk Assessment of Road Tunnels using Bayesian Networks

This paper introduces a methodology for the assessment of risks due to traffic in roadway tunnels. The proposed methodology utilizes Bayesian Probabilistic Networks for the risk assessment. The methodology is based on two hypotheses, i.e. that the roadway tunnel can be represented by a number of homogeneous segments and that the risks associated with a given segment can be represented through a set of Key Performance Indicators (KPI). The KPIs are the observable characteristics of the tunnel and the traffic such as the annual average daily traffic, the number of lanes, possible traffic separation, width, curvature and gradients. The risk is calculated in terms of expected number of accidents, injuries, fatalities and tunnel fires over the entire tunnel length for both, normal traffic events and dangerous goods events. The dependencies and the conditional probabilities in the Bayesian Probabilistic Networks are based on three general pillars, i.e. data, models published in literature and expert opinion. This new approach allows taking into account different specific characteristics of the tunnel and thus the risk of a complex tunnel system can easily be calculated. The Bayesian Probabilistic Networks are embedded into a Microsoft Excel Environment, so that the risk analysis can easily be performed. The methodology is generic, represents the best practice in the field of risk assessment and tunnel research, is realizable and easy to use, is transparent, supports the decision making process, and is modular so that it can be adopted to future findings.

Investigations of Human and Organizational Factors in hazardous vapor accidents

This paper presents a model to assess the contribution of Human and Organizational Factor (HOF) to accidents. The proposed model is made up of two phases. The first phase is the qualitative analysis of HOF responsible for accidents, which utilizes Human Factors Analysis and Classification System (HFACS) to seek out latent HOFs. The hierarchy of HOFs identified in the first phase provides inputs for the analysis in the second phase, which is a quantitative analysis using Bayesian Network (BN). BN enhances the ability of HFACS by allowing investigators or domain experts to measure the degree of relationships among the HOFs. In order to estimate the conditional probabilities of BN, fuzzy analytical hierarchy process and decomposition method are applied in the model. Case studies show that the model is capable of seeking out critical latent human and organizational errors and carrying out quantitative analysis of accidents. Thereafter, corresponding safety prevention measures are derived.

Uncertainty treatment in earthquake modelling using Bayesian probabilistic networks

A probabilistic description of potential ground motion intensity is computed using a Bayesian probabilistic network (BPN) representing the standard probabilistic seismic hazard analysis (PSHA). Two earthquake ground motion intensity parameters are used: response spectral values for structural failures and peak ground acceleration for geotechnical failures. The correlation of these parameters is also considered within a BPN. It is further shown how deaggregation of the seismic hazard could be easily performed using BPNs. A systematic consideration of uncertainty in the values of the parameters of a particular seismic hazard model can be described by PSHA. But the correct choices for elements of the seismic hazard model are uncertain. Logic trees provide a convenient framework for the treatment of model uncertainty. The paper illustrates an alternative way of incorporating the model uncertainty by extending the developed BPN. Incorporation of time-dependent seismic hazard using a BPN is also illustrated. Finally, the uncertainty treatment in earthquake modelling using BPNs is illustrated on the region Adapazari, which is located close to the western part of the North Anatolian Fault in Turkey.

Bayesian probabilistic network approach for managing earthquake risks of cities

This paper considers the application of Bayesian probabilistic networks (BPNs) to large-scale risk based decision making in regard to earthquake risks. A recently developed risk management framework is outlined which utilises Bayesian probabilistic modelling, generic indicator based risk models and geographical information systems. The proposed framework comprises several modules: A module on the probabilistic description of potential future earthquake shaking intensity, a module on the probabilistic assessment of spatial variability of soil liquefaction, a module on damage assessment of buildings and a fourth module on the consequences of an earthquake. Each of these modules is integrated into a BPN. Special attention is given to aggregated risk, i.e. the risk contribution from assets at multiple locations in a city subjected to the same earthquake. The application of the methodology is illustrated on an example considering a portfolio of reinforced concrete structures in a city located close to the western part of the North Anatolian Fault in Turkey.

The potential use of a Gadget model to predict stock responses to climate change in combination with Bayesian networks: the case of Bay of Biscay anchovy

Abstract Andonegi, E., Fernandes, J. A., Quincoces, I., Irigoien, X., Uriarte, A., Pérez, A., Howell, D., and Stefánsson, G. 2011. The potential use of a Gadget model to predict stock responses to climate change in combination with Bayesian networks: the case of Bay of Biscay anchovy. – ICES Journal of Marine Science, 68: 1257–1269. The European anchovy (Engraulis encrasicolus) is a short-lived pelagic species distributed in Atlantic European waters, with the Bay of Biscay being one of the main centres of abundance. Because it is a short-lived species, the state of the stock is determined largely by incoming recruitment. Recruitment is highly variable and depends on a variety of factors, such as the size of the spawning stock and environmental conditions in the area. The use of a coupled model that could serve to predict the evolution of the anchovy stock in the short, medium, and long term under several fishing-pressure scenarios and given climate scenarios is demonstrated. This coupled model consists of a Gadget (Globally Applicable Disaggregated General Ecosystem Toolbox) model that was used to analyse the status of the Bay of Biscay anchovy population and to simulate future scenarios based on the estimated recruitment levels, combined with a probabilistic Bayesian network model for recruitment estimation based on machine-learning methods and using climatic indices as potential forecasting factors. The results indicate that certain combinations of medium to high fishing pressure and adverse environmental conditions could force the stock outside its biological reference boundaries.

Uniqueness of the Level Two Bayesian Network Representing a Probability Distribution

Bayesian Networks are graphic probabilistic models through which we can acquire, capitalize on, and exploit knowledge. they are becoming an important tool for research and applications in artificial intelligence and many other fields in the last decade. This paper presents Bayesian networks and discusses the inference problem in such models. It proposes a statement of the problem and the proposed method to compute probability distributions. It also uses D‐separation for simplifying the computation of probabilities in Bayesian networks. Given a Bayesian network over a family I of random variables, this paper presents a result on the computation of the probability distribution of a subset S of I using separately a computation algorithm and D‐separation properties. It also shows the uniqueness of the obtained result.

Body composition predicted with a Bayesian network from simple variables

The relative contributions of fat-free mass (FFM) and fat mass (FM) to body weight are key indicators for several major public health issues. Predictive models could offer new insights into body composition analysis. A non-parametric equation derived from a probabilistic Bayesian network (BN) was established by including sex, age, body weight and height. We hypothesised that it would be possible to assess the body composition of any subject from easily accessible covariables by selecting an adjusted FFM value within a reference dual-energy X-ray absorptiometry (DXA) measurement database (1999-2004 National Health and Nutrition Examination Survey (NHANES), n 10402). FM was directly calculated as body weight minus FFM. A French DXA database (n 1140) was used (1) to adjust the model parameters (n 380) and (2) to cross-validate the model responses (n 760). French subjects were significantly different from American NHANES subjects with respect to age, weight and FM. Despite this different population context, BN prediction was highly reliable. Correlations between BN predictions and DXA measurements were significant for FFM (R2 0·94, P<0·001, standard error of prediction (SEP) 2·82kg) and the percentage of FM (FM%) (R2 0·81, P<0·001, SEP 3·73%). Two previously published linear models were applied to the subjects of the French database and compared with BN predictions. BN predictions were more accurate for both FFM and FM than those obtained from linear models. In addition, BN prediction generated stochastic variability in the FM% expressed in terms of BMI. The use of such predictions in large populations could be of interest for many public health issues.

Use of Bayesian networks classifiers for long-term mean wind turbine energy output estimation at a potential wind energy conversion site

Due to the interannual variability of wind speed a feasibility analysis for the installation of a Wind Energy Conversion System at a particular site requires estimation of the long-term mean wind turbine energy output. A method is proposed in this paper which, based on probabilistic Bayesian networks (BNs), enables estimation of the long-term mean wind speed histogram for a site where few measurements of the wind resource are available. For this purpose, the proposed method allows the use of multiple reference stations with a long history of wind speed and wind direction measurements. That is to say, the model that is proposed in this paper is able to involve and make use of regional information about the wind resource. With the estimated long-term wind speed histogram and the power curve of a wind turbine it is possible to use the method of bins to determine the long-term mean energy output for that wind turbine. The intelligent system employed, the knowledgebase of which is a joint probability function of all the model variables, uses efficient calculation techniques for conditional probabilities to perform the reasoning. This enables automatic model learning and inference to be performed efficiently based on the available evidence. The proposed model is applied in this paper to wind speeds and wind directions recorded at four weather stations located in the Canary Islands (Spain). Ten years of mean hourly wind speed and direction data are available for these stations. One of the conclusions reached is that the BN with three reference stations gave fewer errors between the real and estimated long-term mean wind turbine energy output than when using two measure–correlate–predict algorithms which were evaluated and which use a linear regression between the candidate station and one reference station.

Modeling the effectiveness of oil combating from an ecological perspective – A Bayesian network for the Gulf of Finland; the Baltic Sea

Maritime traffic poses a major threat to marine ecosystems in the form of oil spills. The Gulf of Finland, the easternmost part of the Baltic Sea, has witnessed a rapid increase in oil transportation during the last 15 years. Should a spill occur, the negative ecological impacts may be reduced by oil combating, the effectiveness of which is, however, strongly dependent on prevailing environmental conditions and available technical resources. This poses increased uncertainty related to ecological consequences of future spills. We developed a probabilistic Bayesian network model that can be used to assess the effectiveness of different oil combating strategies in minimizing the negative effects of oil on six species living in the Gulf of Finland. The model can be used for creating different accident scenarios and assessing the performance of various oil combating actions under uncertainty, which enables its use as a supportive tool in decision-making. While the model is confined to the western Gulf of Finland, the methodology is adaptable to other marine areas facing similar risks and challenges related to oil spills.