China
Africa
Explore chapters and articles related to this topic
Machine Learning
Published in Pedro Larrañaga, David Atienza, Javier Diaz-Rozo, Alberto Ogbechie, Carlos Puerto-Santana, Concha Bielza, Industrial Applications of Machine Learning, 2019
Pedro Larrañaga, David Atienza, Javier Diaz-Rozo, Alberto Ogbechie, Carlos Puerto-Santana, Concha Bielza
For Bayesian networks, we have selected the following five software tools: HUGIN15, GeNIe, Open-Markov16, gRain17 and bnlearn18. HUGIN (Madsen et al., 2005) is a software package –developed by HUGIN EXPERT, a company located in Aalborg, Denmark– for building and deploying decision support systems for reasoning and decision making under uncertainty. HUGIN software is based on Bayesian network and influence diagram technology. The HUGIN software package consists of the HUGIN Decision Engine (HDE), a GUI and application program interfaces (APIs) to facilitate the integration of HUGIN into applications. GeNIe modeler (Druzdzel, 1999) is a GUI providing an interactive model for building and learning Bayesian networks and it is connected with SMILE (Structural Modeling Inference and Learning Engine), which provides exact and approximate inference algorithms. It is based on research at the University of Pittsburgh, USA. Nowadays, it is developed by BayesFusion, LLC. Open-Markov (Arias et al., 2012) is a software tool that implements both constraint-based and score+search learning algorithms and approximate inference methods. It is under development at the National University for Distance Education in Madrid. gRain (Højsgaard, 2012) is an R package for evidence propagation in probabilistic graphical models developed at Aalborg University. bnlearn (Scutari, 2010) is an R package that includes several algorithms for learning the structure and parameters of Bayesian networks from data with either discrete or continuous variables. It implements both constraint-based and score-based algorithms and also includes parallel computing functionalities.
Supply-side risk modelling using Bayesian network approach
Published in Supply Chain Forum: An International Journal, 2022
Satyendra Kumar Sharma, Srikanta Routroy, Udayan Chanda
The Bayesian network was modelled in Hugin Expert software (http://www/hugin.com/) to perform diagnostics and make predictions. Hugin Expert software has been used many other diverse applications in computational biology, healthcare, business simulations. Figure 3 shows the constructed model in the Hugin Expert. The software consolidates the statistical information and knowledge on causal dependencies in the real world, and enables users to specify the distribution of various nodes and parameters. The continuous data of each variable were discretized into five states or classes (i.e., Very Low, Low, Medium, High, and Very High). To train the Bayesian network classifiers, all the attributes are discretised as suggested by Fayyad and Irani. The discretised intervals for all attributes is shown in Appendix. The distribution of variables is shown in Figure 3, which shows that variables have different frequencies in each class. Two variables (nodes; i.e., lead time and inaccurate forecasting) scored over 50% probability. The four risk factors scored the probability considering the two upper classes (i.e., high and very high), such as transport risk: 20.44 + 11.55 = 32.09%; technology risk: 26.02 + 11.11 = 37.13%; inventory risk: 24.06 + 12.72 = 36.78%; and supplier risk: 22.76 + 21.11 = 43.87%. The mean value and standard deviation are provided for each node in Figure 3. Data from experts were fed into the software to run simulation and obtain the results. Figure 3 provides the proposed supply risk model in Hugin Expert.
A Bayesian network model on the interlinkage between Socially Responsible HRM, employee satisfaction, employee commitment and organizational performance
Published in Journal of Management Analytics, 2020
In this paper, the CPTs of the given structure were found by EM algorithm using Hugin Educational 8.5 software (Alameddine et al., 2011). Madsen et al. (2003) discussed the competency of the Hugin tool to learn the structure and parameters of a BN. As discussed to learn the network data from responses a questionnaire of 12 items were considered. Validation was carried out by using two-fold partition (García-Herrero, Mariscal, García-Rodríguez, & Ritzel, 2012). We randomly split the data set into a 70% training set (n = 244) and a 30% validation set (n = 105), with which accuracy was assessed. Validation of the proposed BN model was done on the random subset of 105 cases (30% of the total cases) that were not used earlier. Performance of the proposed BN on training data was validated using the fit criteria like Akaike information criterion (AIC) and Bayesian information criterion (BIC) values. Also to identify the irrelevant or redundant features (variables) during learning of the proposed BN model, Feature Selection Analyzer of the Hugin tool is used. Khor, Ting, and Amnuaisuk (2009) emphasized the importance of identifying the optimal feature set with fewer numbers of features that provide high model accuracy. In Hugin, p-values are used for feature selection to analyze the marginal independence of variables to the target node. The higher the value of “p” the more chances are there that the associated variables are not connected to the target node. We expect, on the evaluation of conditional probability of each attribute, we will be able to identify the key factors that may affect the organizational performance.