Catálogo de publicaciones - libros

Recently, multiagent systems and data mining have attracted considerable attention in the computer science community. This paper combines these two hot research areas to introduce the term on a cooperative learning system, which investigates employing data mining on a cooperative multiagent system. Learning in a partially observable and dynamic multiagent systems environment still constitutes a difficult and major research problem that is worth further investigation. Reinforcement learning has been proposed as a strong method for learning in multi-agent systems. So far, many researchers have proposed various methods to improve the learning ability in multiagent systems. However, reinforcement learning still has some drawbacks. One drawback is not modeling other learning agents present in the domain as part of the state of the environment. Another drawback is that even in learning case, some state-action pairs are experienced much less than others. In order to handle these problems, we describe a new action selection model based on association rules mining. Experimental results obtained on a well-known pursuit domain show the applicability, robustness and effectiveness of the proposed learning approach.

Many business organizations generate a huge amount of transaction data. Association rule mining is a powerful analysis tool to extract the useful meanings and associations from large databases and many automated systems have been developed for mining association rules. However, most of these systems usually mine many association rules from large databases and it is not easy for a user to extract meaningful rules. Visualization has become an important tool in the data mining process for extracting meaningful knowledge and information from large data sets. Though there are several techniques for visualizing mined association rules, most of these techniques visualize the entire set of discovered association rules on a single screen. Such a dense display can overwhelm analysts and reduce their capability of interpretation. In this paper we present a novel technique called for visualizing mined association rules. VisAR consists of four major stages for visualizing mined association rules. These stages include , , , and . Our technique allows an analyst to view only a particular subset of association rules which contain selected items of interest. VisAR is able to display not only many-to-one but also many-to-many association rules. Moreover, our technique can overcome problems of screen clutter and occlusion.

A large number of text files, including HTML documents and XML documents, can be organized as tree structures. One objective of data mining is to discover frequent patterns in them. In this paper, first, we introduce a canonical form of free tree, which is based on the secondly, we present some properties of a closed frequent subtree and a maximal frequent subtree as well as their relationships thirdly, we study a pruning technique of frequent free subtree and improvement on the mining of the nonclosed frequent free subtree; finally, we present an algorithm that mines all closed and maximal frequent free trees and prove validity of this algorithm.

An active research area in Machine Learning is the construction of multiple classifier systems to increase learning accuracy of simple classifiers. In this paper we present E-CIDIM, a multiple classifier system designed to improve the performance of CIDIM, an algorithm that induces small and accurate decision trees. E-CIDIM keeps a maximum number of trees and it induces new trees that may substitute the old trees in the ensemble. The substitution process finishes when none of the new trees improves the accuracy of any of the trees in the ensemble after a pre-configured number of attempts. In this way, the accuracy obtained thanks to an unique instance of CIDIM can be improved. In reference to the accuracy of the generated ensembles, E-CIDIM competes well against bagging and boosting at statistically significance confidence levels and it usually outperforms them in the accuracy and the average size of the trees in the ensemble.

This paper addresses a new classification technique: partially supervised classification (PSC), which is used to identify a specific land-cover class of interest from a remotely sensed image by using unique training samples belong to a specifically selected class. This paper also presents and discusses a novel Support Vector Machine (SVM) algorithm for PSC. Its training set includes labeled samples belong to the class of interest and unlabeled samples of all classes randomly selected from a remotely sensed image. Moreover, all unlabeled samples are assumed to be training samples of other classes and each of them is assigned a weighting factor indicating the likelihood of this assumption; hence, the algorithm is so-called ‘Weighted Unlabeled Sample SVM’ (WUS-SVM). Experimental results with both simulated and real data sets indicate that the proposed PSC method is more robust than 1-SVM and has comparable accuracy to a standard SVM.

Associative classification is a well-known technique which uses association rules to predict the class label for new data object. This model has been recently reported to achieve higher accuracy than traditional classification approaches. There are various strategies for good associative classification in its three main phases: rules generation, rules pruning and classification. Based on a systematic study of these strategies, we propose a new framework named MCRAC, i.e., . MCRAC integrates the advantages of the previously proposed effective strategies as well as the new strategies presented in this paper. An extensive performance study reveals that the advantages of the strategies and the improvement of MCRAC outperform other associative classification approaches on accuracy.

For interactive data mining of very large databases a method working with relatively small training data that can be extracted from the target databases by sampling is proposed, because it takes very long time to generate decision trees for the data mining of very large databases that contain many continues data values, and size of decision trees has the tendency of dependency on the size of training data. The method proposes to use samples of confidence in proper size as the training data to generate comprehensible trees as well as to save time. For medium or small databases direct use of original data with some harsh pruning may be used, because the pruning generates trees of similar size with smaller error rates.

Collaborative filtering is becoming a popular technique for reducing information overload. However, most of current collaborative filtering algorithms have three major limitations: accuracy, data sparsity and scalability. In this paper, we propose a new collaborative filtering algorithm to solve the problem of data sparsity and improve the prediction accuracy. If the rated items amount of a user is less than some threshold, the algorithm utilizes the output of latent class models for neighbors selection, then uses the neighborhood-based method to produce the prediction of unrated items, otherwise it predicts the rating using the STIN1 method. Our experimental results show that our algorithm outperforms the conventional neighborhood-based method and the STIN1 method.

A new polynomial smooth method for solving the support vector machine (SVM) is presented in this paper. It is called the polynomial smooth support vector machine (PSSVM). BFGS method and Newton-Armijo method are applied to solve the PSSVM. Numerical experiments confirm that PSSVM is more effective than SVM.

Knowledge reduction is an important issue in data mining. This paper focuses on the problem of knowledge reduction in incomplete decision tables. Based on a concept of incomplete conditional entropy, a new reduct definition is presented for incomplete decision tables and its properties are analyzed. Compared with several existing reduct definitions, the new definition has a better explanation for knowledge uncertainty and is more convenient for application of the idea of approximate reduct in incomplete decision tables.