Nelson F. F. Ebecken

FuzzyFTA: a fuzzy fault tree system for uncertainty analysis

Abstract This paper describes a new approach and new computational system, FuzzyFTA, for reliability analysis using fault tree and fuzzy logic. Some measures are defined to determine critical components and the uncertainty contribution of each one to the system. The FuzzyFTA system includes algorithms to consider the minimal cut set approach for the top event calculation. After that, these algorithms are used to determine importance measures. The computer code application is the Auxiliary Feedwater System (AFWS) analysis, a recent study made for Angra-I, Brazilian NPP.

EXTRACTING RULES FROM MULTILAYER PERCEPTRONS IN CLASSIFICATION PROBLEMS: A CLUSTERING-BASED APPROACH

Abstract Multilayer perceptrons adjust their internal parameters performing vector mappings from the input to the output space. Although they may achieve high classification accuracy, the knowledge acquired by such neural networks is usually incomprehensible for humans. This fact is a major obstacle in data mining applications, in which ultimately understandable patterns (like classification rules) are very important. Therefore, many algorithms for rule extraction from neural networks have been developed. This work presents a method to extract rules from multilayer perceptrons trained in classification problems. The rule extraction algorithm basically consists of two steps. First, a clustering genetic algorithm is applied to find clusters of hidden unit activation values. Then, classification rules describing these clusters, in relation to the inputs, are generated. The proposed approach is experimentally evaluated in four datasets that are benchmarks for data mining applications and in a real-world meteorological dataset, leading to interesting results.

Evaluating the correlation between objective rule interestingness measures and real human interest

In the last few years, the data mining community has proposed a number of objective rule interestingness measures to select the most interesting rules, out of a large set of discovered rules. However, it should be recalled that objective measures are just an estimate of the true degree of interestingness of a rule to the user, the so-called real human interest. The latter is inherently subjective. Hence, it is not clear how effective, in practice, objective measures are. More precisely, the central question investigated in this paper is: “how effective objective rule interestingness measures are, in the sense of being a good estimate of the true, subjective degree of interestingness of a rule to the user?” This question is investigated by extensive experiments with 11 objective rule interestingness measures across eight real-world data sets.

Bayesian networks for imputation in classification problems

Missing values are an important problem in data mining. In order to tackle this problem in classification tasks, we propose two imputation methods based on Bayesian networks. These methods are evaluated in the context of both prediction and classification tasks. We compare the obtained results with those achieved by classical imputation methods (Expectation–Maximization, Data Augmentation, Decision Trees, and Mean/Mode). Our simulations were performed by means of four datasets (Congressional Voting Records, Mushroom, Wisconsin Breast Cancer and Adult), which are benchmarks for data mining methods. Missing values were simulated in these datasets by means of the elimination of some known values. Thus, it is possible to assess the prediction capability of an imputation method, comparing the original values with the imputed ones. In addition, we propose a methodology to estimate the bias inserted by imputation methods in classification tasks. In this sense, we use four classifiers (One Rule, Naïve Bayes, J4.8 Decision Tree and PART) to evaluate the employed imputation methods in classification scenarios. Computing times consumed to perform imputations are also reported. Simulation results in terms of prediction, classification, and computing times allow us performing several analyses, leading to interesting conclusions. Bayesian networks have shown to be competitive with classical imputation methods.

Towards efficient variables ordering for Bayesian networks classifier

Traditionally, the task of learning Bayesian Networks (BNs) from data has been treated as a NP-Hard search problem. To overcome such difficulty in terms of computational complexity, several approximations have been designed, such as imposing a previous ordering on the domain attributes that restrict the number of Bayesian structures to be learned or using other approaches trying to reduce the state space of this problem. In this paper, we propose a simple method based on feature ranking algorithms which has low computational complexity (O(n^2), where n is the number of variables) and produces good results. We empirically demonstrate that feature ranking algorithms (namely, Chi-Squared and Information Gain) can be used to define efficient variables ordering in the BNC learning context. The proposed method can bring improvements, when using the K2 algorithm, to learn a Bayesian Network Classifier from data.

Knowledge discovering for coastal waters classification

Since almost all anthropogenic activities ultimately affect the coastal waters, access properties and processes in this environment is the major issue in decision making and system management. Particularly, seasonal patterns are not clear in tropical areas, therefore, requiring environmental classification. The knowledge of long-term biogenic element dynamics, the biological response, and the selection of indicators connecting lower and higher trophic levels have became a real need for the sustainable management of marine resources. Under this scenario, this paper uses a machine-learning approach to determine the ecological status of coastal waters based on patterns of occurrence of meroplankton larvae of epibenthic fauna and its relationship with other environmental variables. The case studied is the upwelling influenced bay at Cabo Frio Island (Rio de Janeiro - Brazil) because this location has been suffering with anthropogenic impact. Models of crisp and fuzzy rules have been tested as classifiers. Results show it is possible to access hidden patterns of water masses within a set of association rules.

Growing compact RBF networks using a genetic algorithm

A novel approach for applying genetic algorithms to the configuration of radial basis function networks is presented. A new crossover operator that allows for some control over the competing conventions problem is introduced. Also, a minimalist initialization scheme which tends to generate more parsimonious models is also presented. Finally, a reformulation of generalized cross-validation criterion for model selection, making it more conservative, is discussed. The proposed model is submitted to a computational experiment in order to verify its effectiveness.

Feature selection by Bayesian networks

This work both describes and evaluates a Bayesian feature selection approach for classification problems. Basically, a Bayesian network is generated from a dataset, and then the Markov Blanket of the class variable is used to the feature subset selection task. The proposed methodology is illustrated by means of simulations in three datasets that are benchmarks for data mining methods: Wisconsin Breast Cancer, Mushroom and Congressional Voting Records. Three classifiers were employed to show the efficacy of the proposed method. The average classification rates obtained in the datasets formed by all features are compared to those achieved in the datasets formed by the features that belong to the Markov Blanket. The performed simulations lead to interesting results.

A comparison of models for uncertainty analysis by the finite element method

Abstract Uncertainty in structural engineering analysis exists in the architecture of a structural system, its basic parameters, the information resulting from the abstracted aspects of the system, and the non-abstracted or unknown aspects of the system. Also, uncertainty is present as a result of prediction models, analysis and design of structures, and general lack of knowledge about the behavior of real structures. One of the important factors that lead to errors in numerical predictions is the degree of precision in obtaining the relevant parameters. In this paper we discuss two different methodologies: 1. Classical probabilistic approach, in which the properties are treated as random variables. Stochastic Finite Element Methods are examined using both Monte Carlo Simulation and Perturbation Methods. 2. Possibilistic approach, by a model based on the theory of fuzzy sets. Some results are presented to point out the main characteristics of the two methodologies (Lima, D.Sc. Thesis, COPPE/Federal University Rio de Janeiro, 1996).

Evaluating a Nearest-Neighbor Method to Substitute Continuous Missing Values

This work proposes and evaluates a Nearest-Neighbor Method to substitute missing values in datasets formed by continuous attributes. In the substitution process, each instance containing missing values is compared with complete instances, and the closest instance is used to assign the attribute missing value. We evaluate this method in simulations performed in four datasets that are usually employed as benchmarks for data mining methods - Iris Plants, Wisconsin Breast Cancer, Pima Indians Diabetes and Wine Recognition. First, we con- sider the substitution process as a prediction task. In this sense, we em- ploy two metrics (Euclidean and Manhattan) to simulate substitutions both in original and normalized datasets. The obtained results were compared to those provided by a usually employed method to perform this task, i.e. substitution by the mean value. Based on these simulations, we propose a substitution procedure for the well-known K-Means Clustering Algorithm. Then, we perform clustering simulations, com- paring the results obtained in the original datasets with the substituted ones. These results indicate that the proposed method is a suitable esti- mator for substituting missing values, i.e. it preserves the relationships between variables in the clustering process. Therefore, the proposed Nearest-Neighbor Method is an appropriate data preparation tool for the K-Means Clustering Algorithm.