Catálogo de publicaciones - libros

In this paper, we propose an efficient palmprint recognition scheme which has two features: 1) representation of palm images by two dimensional (2-D) wavelet subband coefficients and 2) recognition by a modular, personalized classification method based on Kernel Principal Component Analysis (Kernel PCA). Wavelet subband coefficients can effectively capture substantial palm features while keeping computational complexity low. We then kernel transforms to each possible training palm samples and then mapped the high-dimensional feature space back to input space. Weighted Euclidean linear distance based nearest neighbor classifier is finally employed for recognition. We carried out extensive experiments on PolyU Palmprint database includes 7752 palms from 386 different palms. Detailed comparisons with earlier published results are provided and our proposed method offers better recognition accuracy (99.654%).

Interactive data mining focuses on efficient and effective human-computer interactions for data analysis purposes. An interactive system is an integration of a human user and a computer machine. ICS, an interactive classification system, is implemented to demonstrate the power of interactive data mining. The interaction is mutually beneficial to users and machines. This article describes the architecture of ICS, and introduces the main features of ICS in the entire data mining process.

The nearest neighbor () classifier has been a widely used technique in pattern recognition because of its simplicity and good behavior. To decide the class of a new object, the classifier performs an exhaustive comparison between the object to classify and the training set . However, when is large, the exhaustive comparison is very expensive and sometimes becomes inapplicable. To avoid this problem, many fast algorithms have been developed for numerical object descriptions, most of them based on metric properties to avoid comparisons. However, in some sciences as Medicine, Geology, Sociology, etc., objects are usually described by numerical and non numerical attributes (mixed data). In this case, we can not assume the comparison function satisfies metric properties. Therefore, in this paper a fast most similar object classifier based on search methods suitable for mixed data is presented. Some experiments using standard databases and a comparison with other two fast methods are presented.

This study emphasizes the importance of using appropriate measures in particular text classification settings. We focus on methods that evaluate how well a classifier performs. The effect of transformations on the confusion matrix are considered for eleven well-known and recently introduced classification measures. We analyze the measure’s ability to retain its value under changes in a confusion matrix. We discuss benefits from the use of the invariant and non-invariant measures with respect to characteristics of data classes.

This paper explores two simple and efficient pre-pruning strategies for the cost-sensitive decision tree algorithm to avoid overfitting. One is to limit the cost-sensitive decision trees to a depth of two. The other is to prune the trees with a pre-specified threshold. Empirical study shows that, compared to the error-based tree algorithm C4.5 and several other cost-sensitive tree algorithms, the new cost-sensitive decision trees with pre-pruning are more efficient and perform well on most UCI data sets.

Locally weighted naive Bayes (LWNB) is a successful instance-based classifier, which first finds the neighbors of the test instance using Euclidean metric, and then builds a naive Bayes model in the local neighborhood. However, Euclidean metric is not the best choice for LWNB. For nominal attributes, Euclidean metric has to order and number the values of attributes, or judge whether the attribute values are identical or not. For numeric attributes, Euclidean metric is not appropriate for different attribute scales and variability, and encounters the problem of attribute value outliers when normalizing values. In this paper, we systematically study probability based metrics, such as Interpolated Value Difference Metric (IVDM), Extended Short and Fukunaga Metric (SF2), SF2 calibrated by logarithm (SF2LOG) and Minimum Risk Metric (MRM), and apply them to LWNB. These probability based metrics can solve the above problems of Euclidean metric since they depend on the difference between the probabilities to evaluate the distances between the instances. We conduct the experiments to compare the performances of LWNB classifiers using Euclidean metric and probability based metrics on UCI datasets. The results show that LWNB classifiers using IVDM outperform the ones using Euclidean metric and other probability based metrics. We also observe that SF2, SF2LOG and MRM do not perform well due to their inaccurate probability estimates. An artificial dataset is built by logical sampling in a Bayesian network, where accurate probability estimates can be produced. We conduct the experiment on the artificial dataset. The results show that SF2, SF2LOG and MRM using accurate probability estimates perform better than Euclidean metric and IVDM in LWNB.

Boosted ensemble classifiers have a demonstrated ability to discover regularities in large, poorly modeled datasets. In this paper we present an application of multi-hypothesis AdaBoost to detect epileptiform activity from electrophysiological recordings. While existing boosting methods do not account automatically for the sequence information that is available when analyzing time-series data, we present a recurrent extension to AdaBoost, and show that it improves classification accuracy in our application domain.

We have constructed a poker classification system which makes informed betting decisions based upon three defining features extracted while playing poker: hand value, risk, and aggressiveness. The system is implemented as a player-agent, therefore the goals of the classifier are not only to correctly determine whether each hand should be folded, called, or raised, but to win as many chips as possible from the other players. The decision space is found by evolutionary methods, starting from a data-driven initial state. Our results showed that evolving an agent from a data-driven “head-start” position resulted in the best performance over agents evolved from scratch, data-driven agents, random agents, and “always fold” agents.

and have been known in the CSP discipline for fifteen years, especially in scheduling, planning, diagnosis and configuration domains. Basically a restricts the participation of a variable in a feasible scenario while a composite variable allows us to express a disjunction of variables or sub CSPs where only one will be added to the problem to solve. In this paper we combine the features of and in a unique framework that we call . Our framework allows the representation of dynamic constraint problems where all the information corresponding to any possible change are available a priori. Indeed these latter information are added to the problem to solve in a dynamic manner, during the resolution process, via conditional (or activity) constraints and composite variables. A composite variable is a variable whose possible values are CSP variables. In other words this allows us to represent disjunctive variables where only one will be added to the problem to solve. An activity constraint activates a non active variable (this latter variable will be added to the problem to solve) if a given condition holds on some other active variables. In order to solve the CCCSP, we propose two methods that are respectively based on constraint propagation and Stochastic Local Search (SLS). The experimental study, we conducted on randomly generated CCCSPs demonstrates the efficiency of a variant of the MAC strategy (that we call MAC+) over the other constraint propagation techniques. We will also show that MAC+ outperforms the SLS method MCRW for highly consistent CCCSPs. MCRW is however the procedure of choice for under constrained and middle constrained problems and also for highly constrained problems if we trade search time for the quality of the solution returned (number of solved constraints).

Variables and constraints in problem domains are often distributed. These distributed constraint satisfaction problems (DCSPs) lend themselves to multiagent solutions. Most existing algorithms for DCSPs are extensions of centralized backtracking or iterative improvement with breakout. Their worst case complexity is exponential. On the other hand, directional consistency based algorithms solve centralized CSPs efficiently if primal graph density is bounded. No known multiagent algorithms solve DCSPs with the same efficiency. We propose the first such algorithm and show that it is sound and complete.