## Data Analysis with Bayesian Networks: A Bootstrap Approach (1999)

### Cached

### Download Links

Citations: | 48 - 7 self |

### BibTeX

@MISC{Friedman99dataanalysis,

author = {Nir Friedman and Moises Goldszmidt and Abraham Wyner},

title = {Data Analysis with Bayesian Networks: A Bootstrap Approach},

year = {1999}

}

### Years of Citing Articles

### OpenURL

### Abstract

In recent years there has been significant progress in algorithms and methods for inducing Bayesian networks from data. However, in complex data analysis problems, we need to go beyond being satisfied with inducing networks with high scores. We need to provide confidence measures on features of these networks: Is the existence of an edge between two nodes warranted? Is the Markov blanket of a given node robust? Can we say something about the ordering of the variables? We should be able to address these questions, even when the amount of data is not enough to induce a high scoring network. In this paper we propose Efron's Bootstrap as a computationally efficient approach for answering these questions. In addition, we propose to use these confidence measures to induce better structures from the data, and to detect the presence of latent variables.

### Citations

2562 |
An Introduction to the Bootstrap
- Efron, Tibshirani
- 1993
(Show Context)
Citation Context ...n determine the level of confidence about various structural features of the Bayesian networks we induce from data sets. We consider an approach and methodology based on the Bootstrap method of Efron =-=[7]-=- for addressing this type of challenges. The Bootstrap is a computer-based method for assigning measures of accuracy to statistics estimates and performing statistical inference. We regard these measu... |

1138 |
Confidence limits on phylogenies: an approach using the bootstrap
- Felsenstein
- 1985
(Show Context)
Citation Context ... be concisely defined in a simple expression (see [7]). An example is the application of the bootstrap in evolutionary biology to measure confidence in inferences from phylogenetic trees. Felsenstein =-=[9]-=-, has applied re-sampling tools to estimate uncertainty in edges (clades) of evolutionary trees (which specify the phylogenetic evolution of a gene over time). Similar to phylogenies, we test re-sampl... |

909 | Learning Bayesian networks: The combination of knowledge and statistical data
- Heckerman, Geiger, et al.
- 1995
(Show Context)
Citation Context ... (or a score) that evaluates the “fitness” of networks with respect to the training data, and then to search for the best network (according to this score). In this paper we use the score proposed in =-=[13]-=- which is based on Bayesian considerations, and which scores a network structure according to the posterior probability of the graph structure given the training data (up to a constant). We note that ... |

862 | A tutorial on learning with Bayesian Networks
- Heckerman
- 1995
(Show Context)
Citation Context ...ter structures from the data, and to detect the presence of latent variables. 1 Introduction In the last decade there has been a great deal of research focused on learning Bayesian networks from data =-=[2, 12]-=-. With few exceptions, these results have concentrated on computationally efficient induction methods and, more recently, on the issue of hidden variables and missing data. The main concern in this li... |

790 |
Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridisation
- Spellman, Sherlock, et al.
- 1998
(Show Context)
Citation Context ...andom variables and � 46 edges, only 4 of which are undirected in the PDAG. This is a standard benchmark in the learning literature. ������� . A network induced using a gene expression � dataset from =-=[8]-=- for 76 genes. Genes were grouped by a clustering algorithm that searches for groups of related genes (details of the induction can be found in [11]). The network has 140 edges, only 5 of which are un... |

619 |
Exploring the metabolic and genetic control of gene expression on a genomic scale
- DeRisi, Iyer, et al.
- 1997
(Show Context)
Citation Context ... �� ����� in turn translated into a protein molecule. Recent technical breakthroughs in molecular biology enable biologists to measure of the expression levels of thousands of genes in one experiment =-=[6, 17, 21]-=-. The data generated from these experiments consists of instances, each one of which has thousands of attributes. However, the largest datasets available today contain only few hundreds of instances. ... |

205 | A Theory of Inferred Causation
- Pearl, Verma
- 1991
(Show Context)
Citation Context ...it. Two Bayesian network � structures ��� and are equivalent, if they imply exactly the same set of independence statements. The characterization of Bayesian network equivalence classes is studied in =-=[3, 18, 19, 20]-=-. Results in these papers establish that equivalent networks agree on the connectivity between variables, but might disagree on the direction of the arcs. These results also show that each equivalence... |

171 | A guide to the literature on learning probabilistic networks from data
- Buntine
- 1996
(Show Context)
Citation Context ...ter structures from the data, and to detect the presence of latent variables. 1 Introduction In the last decade there has been a great deal of research focused on learning Bayesian networks from data =-=[2, 12]-=-. With few exceptions, these results have concentrated on computationally efficient induction methods and, more recently, on the issue of hidden variables and missing data. The main concern in this li... |

156 | Learning Bayesian networks is NP-complete
- Chickering
- 1996
(Show Context)
Citation Context ...duced indirectly, since presumably the “right” structure is the one that can better generalize from the training data. Finding the structure that maximizes the score is usually an intractable problem =-=[4]-=-. Thus, we usually resort to heuristic search to find a high-scoring structure. Standard proposals for such search include greedy hill-climbing, stochastic hill-climbing, and simulated annealing; see ... |

144 |
Large-scale temporal gene expression mapping of central nervous system development
- Wen, Fuhrman, et al.
- 1998
(Show Context)
Citation Context ... �� ����� in turn translated into a protein molecule. Recent technical breakthroughs in molecular biology enable biologists to measure of the expression levels of thousands of genes in one experiment =-=[6, 17, 21]-=-. The data generated from these experiments consists of instances, each one of which has thousands of attributes. However, the largest datasets available today contain only few hundreds of instances. ... |

90 | A transformational characterization of equivalent Bayesian network structures
- Chickering
- 1995
(Show Context)
Citation Context ...it. Two Bayesian network � structures ��� and are equivalent, if they imply exactly the same set of independence statements. The characterization of Bayesian network equivalence classes is studied in =-=[3, 18, 19, 20]-=-. Results in these papers establish that equivalent networks agree on the connectivity between variables, but might disagree on the direction of the arcs. These results also show that each equivalence... |

82 |
Causal inference and causal explanation with background knowledge
- Meek
- 1995
(Show Context)
Citation Context ...it. Two Bayesian network � structures ��� and are equivalent, if they imply exactly the same set of independence statements. The characterization of Bayesian network equivalence classes is studied in =-=[3, 18, 19, 20]-=-. Results in these papers establish that equivalent networks agree on the connectivity between variables, but might disagree on the direction of the arcs. These results also show that each equivalence... |

80 | A Bayesian approach to causal discovery - Heckerman, Meek, et al. - 1999 |

52 |
Array of hope
- Lander
- 1999
(Show Context)
Citation Context ...learly contain valuable information. For example, we would like to induce correlation and causation relations among genes (e.g., high expression levels of one gene “cause” the suppression of another) =-=[16]-=-. The challenge is then, to separate the measurable “signal” in this data from the “noise,” that is, the genuine correlations and causations properties from spurious (random) correlations. Analysis of... |

25 |
The ALARM monitoring system
- Beinlich, Suermondt, et al.
- 1989
(Show Context)
Citation Context ...es the bootstrap estimates depend on various parameters, such as size of dataset, type of feature, and bootstrap method. 4.1 Methodology We performed simulation results from three networks: ��������� =-=[1]-=-. This network has 37 random variables and � 46 edges, only 4 of which are undirected in the PDAG. This is a standard benchmark in the learning literature. ������� . A network induced using a gene exp... |

21 |
Sequential model criticism in probabilistic expert systems
- Cowell, Dawid, et al.
- 1993
(Show Context)
Citation Context ...in of study from features in the induced Bayesian network. The study of methods for establishing the quality of induced Bayesian networks has not been totally ignored in the literature. Cowell et al. =-=[5]-=- present a method based on the log-loss scoring function to monitor each variable in a given network. These monitors check the deviation of the predictions by these variables from the observations in ... |

15 |
Expression monitoring by hybridization to high density oligonucleotide arrays
- Lockhart, Dong, et al.
- 1996
(Show Context)
Citation Context ... �� ����� in turn translated into a protein molecule. Recent technical breakthroughs in molecular biology enable biologists to measure of the expression levels of thousands of genes in one experiment =-=[6, 17, 21]-=-. The data generated from these experiments consists of instances, each one of which has thousands of attributes. However, the largest datasets available today contain only few hundreds of instances. ... |

9 | On the application of the bootstrap for computing confidence measures on features of induced bayesian networks
- Friedman, Goldszmidt, et al.
- 1999
(Show Context)
Citation Context ...chnique towards a given feature. This latter idea nicely separates the variation in the data from the shortcomings of the algorithm. It is this latter interpretation of confidence that was pursued in =-=[10]-=-. The methods introduced in this paper encompass both types of confidence, and focuses on the former (more below). Although the Bootstrap is conceptually easy to implement and apply in our context, th... |

2 |
On using bayesian networks to analyze whole-genome expression data
- Friedman, Nachman, et al.
- 1999
(Show Context)
Citation Context ...twork induced using a gene expression � dataset from [8] for 76 genes. Genes were grouped by a clustering algorithm that searches for groups of related genes (details of the induction can be found in =-=[11]-=-). The network has 140 edges, only 5 of which are undirected in the PDAG. ��� � . A network induced from a dataset of messages ��� from 20 newsgroup [15]. Each document is represented as an instance w... |

2 |
A probabilistic analysis of the rocchio algorithm with tfidf for text categorization
- Jochims
- 1997
(Show Context)
Citation Context ...enes (details of the induction can be found in [11]). The network has 140 edges, only 5 of which are undirected in the PDAG. ��� � . A network induced from a dataset of messages ��� from 20 newsgroup =-=[15]-=-. Each document is represented as an instance with a variable denoting the newsgroup, and 99 boolean variables corresponding to most frequent words (other than stop words) and denoting whether the wor... |