Silvia Acid

Searching for Bayesian network structures in the space of restricted acyclic partially directed graphs

Although many algorithms have been designed to construct Bayesian network structures using different approaches and principles, they all employ only two methods: those based on independence criteria, and those based on a scoring function and a search procedure (although some methods combine the two). Within the score+search paradigm, the dominant approach uses local search methods in the space of directed acyclic graphs (DAGs), where the usual choices for defining the elementary modifications (local changes) that can be applied are arc addition, arc deletion, and arc reversal. In this paper, we propose a new local search method that uses a different search space, and which takes account of the concept of equivalence between network structures: restricted acyclic partially directed graphs (RPDAGs). In this way, the number of different configurations of the search space is reduced, thus improving efficiency. Moreover, although the final result must necessarily be a local optimum given the nature of the search method, the topology of the new search space, which avoids making early decisions about the directions of the arcs, may help to find better local optima than those obtained by searching in the DAG space. Detailed results of the evaluation of the proposed search method on several test problems, including the well-known Alarm Monitoring System, are also presented.

A comparison of learning algorithms for Bayesian networks: a case study based on data from an emergency medical service

Due to the uncertainty of many of the factors that influence the performance of an emergency medical service, we propose using Bayesian networks to model this kind of system. We use different algorithms for learning Bayesian networks in order to build several models, from the hospital managers point of view, and apply them to the specific case of the emergency service of a Spanish hospital. This first study of a real problem includes preliminary data processing, the experiments carried out, the comparison of the algorithms from different perspectives, and some potential uses of Bayesian networks for management problems in the health service.

A hybrid methodology for learning belief networks: BENEDICT

Abstract Previous algorithms for the construction of belief networks structures from data are mainly based either on independence criteria or on scoring metrics. The aim of this paper is to present a hybrid methodology that is a combination of these two approaches, which benefits from characteristics of each one, and to develop two operative algorithms based on this methodology. Results of the evaluation of the algorithms on the well-known Alarm network are presented, as well as the algorithms performance issues and some open problems.

Learning Bayesian Network Classifiers: Searching in a Space of Partially Directed Acyclic Graphs

There is a commonly held opinion that the algorithms for learning unrestricted types of Bayesian networks, especially those based on the score+search paradigm, are not suitable for building competitive Bayesian network-based classifiers. Several specialized algorithms that carry out the search into different types of directed acyclic graph (DAG) topologies have since been developed, most of these being extensions (using augmenting arcs) or modifications of the Naive Bayes basic topology. In this paper, we present a new algorithm to induce classifiers based on Bayesian networks which obtains excellent results even when standard scoring functions are used. The method performs a simple local search in a space unlike unrestricted or augmented DAGs. Our search space consists of a type of partially directed acyclic graph (PDAG) which combines two concepts of DAG equivalence: classification equivalence and independence equivalence. The results of exhaustive experimentation indicate that the proposed method can compete with state-of-the-art algorithms for classification.

An Information Retrieval Model Based on Simple Bayesian Networks

In this article a new probabilistic information retrieval (IR) model, based on Bayesian networks (BNs), is proposed. We first consider a basic model, which represents only direct relationships between the documents in the collection and the terms or keywords used to index them. Next, we study two versions of an extended model, which also represents direct relationships between documents. In either case the BNs are used to compute efficiently, by means of a new and exact propagation algorithm, the posterior probabilities of relevance of the documents in the collection given a query. The performance of the proposed retrieval models is tested through a series of experiments with several standard document collections.

Approximations of Causal Networks by Polytrees: an Empirical Study

Once causal networks have been chosen as the model of knowledge representation of our interest, the aim of this work is to assess the performance of polytrees or Singly connected Causal Networks (SCNs) as approximations of general Multiply connected Causal Networks (MCNs). To do that we have carried out a simulation experiment in which we generated a number of MCNs, simulated them to get samples and used these samples to learn the SCNs that approximated the original MCNs, reporting the results.

Learning with CASTLE

We will describe here the learning algorithms we have implemented in CASTLE, (Causal Structures From Inductive Learning), to learn about causal structures from examples. A brief introduction to the software ifself and a description of what we intend to develop and implement in CASTLE are also given. Finally, the use of CASTLE is illustrated on a simple example.

The Search of Causal Orderings: A Short Cut for Learning Belief Networks

Although we can build a belief network starting from any ordering of its variables, its structure depends heavily on the ordering being selected: the topology of the network, and therefore the number of conditional independence relationships that may be explicitly represented can vary greatly from one ordering to another. We develop an algorithm for learning belief networks composed of two main subprocesses: (a) an algorithm that estimates a causal ordering and (b) an algorithm for learning a belief network given the previous ordering, each one working over different search spaces, the ordering and dag space respectively.

Score-based methods for learning Markov boundaries by searching in constrained spaces

Within probabilistic classification problems, learning the Markov boundary of the class variable consists in the optimal approach for feature subset selection. In this paper we propose two algorithms that learn the Markov boundary of a selected variable. These algorithms are based on the score+search paradigm for learning Bayesian networks. Both algorithms use standard scoring functions but they perform the search in constrained spaces of class-focused directed acyclic graphs, going through the space by means of operators adapted for the problem. The algorithms have been validated experimentally by using a wide spectrum of databases, and their results show a performance competitive with the state-of-the-art.

Evaluation methods and strategies for the interactive use of classifiers

We consider the scenario in which an automatic classifier (previously built) is available. It is used to classify new instances but, in some cases, the classifier may request the intervention of a human (the oracle), who gives it the correct class. In this scenario, first it is necessary to study how the performance of the system should be evaluated, as it cannot be based solely on the predictive accuracy obtained by the classifier but it should also take into account the cost of the human intervention; second, studying the concrete circumstances under which the classifier decides to query the oracle is also important. In this paper we study these two questions and include also an experimental evaluation of the different proposed alternatives.