An Overview of Bayesian Network Applications in Uncertain Domains

Abstract

Abstract-Uncertainty is a major barrier in knowledge discovery from complex problem domains. Knowledge discovery in such domains requires qualitative rather than quantitative analysis. Therefore, the quantitative measures can be used to represent uncertainty with the integration of various models. The Bayesian Network (BN) is a widely applied technique for characterization and analysis of uncertainty in real world domains. Thus, the real application of BN can be observed in a broad range of domains such as image processing, decision making, system reliability estimation and PPDM (Privacy Preserving in Data Mining) in association rule mining and medical domain analysis. BN techniques can be used in these domains for prediction and decision support. In this article, a discussion on general BN representation, draw inferences, learning and prediction is followed by applications of BN in some specific domains. Domain specific BN representation, inferences and learning process are also presented. Building upon the knowledge presented, some future research directions are also highlighted.Index Terms-Uncertainty, knowledge discovery, Bayesian network, image processing, decision making, privacy preservation, system reliability estimation. I. INTRODUCTIONUncertainty is a commonly faced problem in real world applications. Uncertainty can be described as an inadequate amount of information [1]. Nevertheless, uncertainty may also exist in situations that have enough amount of information [2]. Furthermore, uncertainty may be alleviated or eliminated with the addition of new information. Addition of more information in complex processes may lead to mining of limited knowledge. Uncertainty can be computed mathematically with probability theory. In uncertain situations, there is an involvement of possibility of states of attributes. Consequently, the models established on probabilistic inferences have the capability to assign a probabilistic value according to a defined principle. Accordingly, the prediction with large number of states in a model is accomplished. The question rises "how prediction is realized in the presence of large number of states in a model?" An answer to this question is the employment of Bayesian Network (BN) with several variables [3]- [5]. BNs, also known as belief networks, belong to the family of probabilistic graphical models. These graphical structures correspond to knowledge about an uncertain domain. More specifically, each node in the graphical structure represents a random variable, while the edges/arcs between the nodes represent conditional dependencies among nodes. These conditional dependencies are estimated by using acknowledged statistical and computational methods. Consequently, BNs incorporate concepts from graph and probability theory, computer science, and statistics.Since last two decades, BN is recognized as an important tool for a number of expert systems especially in domains involving uncertainty [6]. This recognition of BN has several reasons behind it. First, BN encodes the depen...

An Analysis for the Qualitative Improvement of Education and Learning based on the Way of Learner Errors in Descriptive Questions

“…Regarding the Bayesian networks, refer to Scutari and Denis (2014), Lqbal, Yin, Hao, IIyas and Ali (2015).…”

A System of Serial Computation for Classified Rules Prediction in Non-Regular Ontology Trees

“…Similarly, joint probability distribution algorithm have also been applied in decision support inferencing (e.g. [7], [9], [20], [10], but to the best of our knowledge there is none for the estimation and prediction of the actual number of required probabilistic induction rules or classification learning based on semantic ontology trees.We believe that when the required number of rules are known at the knowledge acquisition stage, then classification learning can be modelled easily without missing any classification category whether manually or automatically. This paper thus presents a system of polynomial equation for estimating and predicting the number of probabilistic inference or induction rules.…”

An industrial Fault Diagnosis System based on Bayesian Networks

“…Those collected data will be used after for doing the learning phase to some algorithms that will predict the researched behavior of the system [15] [4]. Though, the acquisition and the availability of data to be used for ML formalisms such as Artificial Neural Networks and Bayesian Networks are widely posed problems [13] [11]. This paper's approach consist of using one of the latest methods in the modeling domain: A data-oriented method.…”