Search

In recent years, Unsupervised Feature Selection (UFS) methods have attracted considerable interest in different research areas due to their wide application in problems where unlabeled data appears. Nevertheless, few UFS methods can process datasets described by both numerical and non-numerical features (mixed data). This paper introduces a new filter UFS method based on a new feature correlation measure for mixed data to select a relevant and non-redundant feature subset. The proposed method addresses the feature selection problem in two stages through a strategy that combines Spectral Feature Selection to identify relevant features and a Pair-wise Redundancy Analysis to remove those features with a high correlation with others. According to our experiments, the proposed feature selection method based on the introduced feature correlation measure provides better quality results than previous filter UFS methods for mixed data reported in the literature. • A filter Unsupervised Feature Selection method for mixed data is presented. • A new feature correlation measure for mixed data is introduced. • Spectral Feature Selection and Pair-wise Redundancy Analysis were applied. • A comprehensive evaluation of the proposed method is provided. [ABSTRACT FROM AUTHOR]

In supervised classification, a training set is given to a classifier to learn a decision rule for classifying unseen cases. When large training sets are processed, the training stage becomes slow especially for instance-based learning. However, not all information in a training set is useful for classification because it could contain either redundant or noisy prototypes. Therefore a process for discarding useless prototypes is required; this process is known as prototype selection. In this work, we present some methods for selecting prototypes based on prototype relevance, which are accurate and fast for large datasets; in addition, our methods can be applied over datasets described by nominal features. We report experimental results showing the effectiveness of our methods as well as a comparison against other successful prototype selection methods. [ABSTRACT FROM AUTHOR]

Most algorithms to mine graph patterns, during the searching process, require a pattern to be identical to its occurrences, relying on the graph isomorphism problem. However, in recent years, there has been interest in the case in which it is acceptable to have some differences between a pattern and its occurrences, whether these differences are in labels or in structure. Allowing some differences and using inexact matching to measure the similarity between graphs lead to the discovery of new patterns, but some important challenges, such as the increment on the number of found patterns, make the post-mining analysis harder. In this work we focus on two extensions of the AGraP algorithm, which mines inexact patterns, addressing the issue of reducing the output pattern set while trying to retain the useful information gained through the use of inexact matching. First, exploring a traditional approach, we propose the CloseAFG algorithm that focuses on closed patterns. Then, we propose the IntAFG algorithm to find a subset of patterns covering the original pattern set, while lessening redundancy among selected patterns. We show the performance of our approaches through some experiments on synthetic databases; additionally, we also show the usefulness of the reduced pattern sets for image classification. [ABSTRACT FROM AUTHOR]

Feature selection is a common task in areas such as Pattern Recognition, Data Mining, and Machine Learning since it can help to improve prediction quality, reduce computation time and build more understandable models. Although feature selection for supervised classification has been widely studied, feature selection in the absence of class labels, namely feature selection for clustering or unsupervised feature selection, has been less addressed. Most existing unsupervised feature selection approaches suffer from the called “Bias of Criterion Values to Dimension,” which arises when feature subsets with different cardinality are evaluated by an internal evaluation clustering criterion. In this paper, we introduce a new hybrid filter–wrapper method for clustering, which combines the spectral feature selection framework using the Laplacian Score ranking and a modified Calinski–Harabasz index. The proposed method in the filter stage sorts the features according to their relevance, while in the wrapper stage, through our modified Calinski–Harabasz index that takes into account the cardinality of the feature subsets under evaluation, evaluates the features considering them as a subset rather than individually by using two well-known selection strategies. Experiments on different datasets show that the proposed method alleviates the “Bias of Criterion Values to Dimension” and, identifies and selects more relevant features than those selected by other reported hybrid filter–wrapper feature selection methods for clustering. Additionally, we also contrast our results against other filter and wrapper methods of the state-of-the-art. [ABSTRACT FROM AUTHOR]

Abstract: Instance selection algorithms are used for reducing the number of training instances. However, most of them suffer from long runtimes which results in the incapability to be used with large datasets. In this work, we introduce an Instance Ranking per class using Borders (instances near to instances belonging to different classes), using this ranking we propose an instance selection algorithm (IRB). We evaluated the proposed algorithm using k-NN with small and large datasets, comparing it against state of the art instance selection algorithms. In our experiments, for large datasets IRB has the best compromise between time and accuracy. We also tested our algorithm using SVM, LWLR and C4.5 classifiers, in all cases the selection computed by our algorithm obtained the best accuracies in average. [Copyright &y& Elsevier]

• A supervised filter feature selection method for mixed data is introduced • Spectral analysis and Information-theory based feature selection was applied • Relevant and non-redundant features are selected • An evaluation of the proposed method on 40 public datasets was performed • A comprehensive comparison against state-of-the-art methods was made Spectral analysis and Information-theory are two powerful and successful frameworks for feature selection in supervised classification problems. However, most of the methods developed under these frameworks have been introduced for handling exclusively numerical or non- numerical data. In this paper, we propose a supervised filter feature selection method that combines Spectral Feature Selection and Information-theory based redundancy analysis for selecting relevant and non-redundant features in supervised mixed datasets; i.e., datasets where the objects are described simultaneously by both, numerical and non-numerical features. To demonstrate the effectiveness of our proposed supervised filter feature selection method, we conducted several experiments on 40 public real-world datasets. Additionally, we compare our method against relevant state-of-the-art supervised filter methods for numerical, non-numerical, and mixed data. From this comparison, our method, in general, obtains better results than the results obtained by the other evaluated filter feature selection methods. [ABSTRACT FROM AUTHOR]

• A new algorithm for computing all the shortest reducts of a decision system is introduced. • Some of the most effective pruning properties for computing all reducts are adapted to compute only the shortest reducts. • Our algorithm uses binary cumulative operations and a fast candidate evaluation process unlike other algorithms. • Our proposal is the fastest in decision systems whose simplified binary discernibility matrices have medium or high density. This paper deals with the problem of computing the shortest reducts of a decision system. The shortest reducts are useful for attribute reduction in classification problems and data size reduction. Unfortunately, finding all the shortest reducts is an NP-hard problem. There are some algorithms reported in the literature to overcome the complexity of computing the shortest reducts. However, most of these algorithms relay on costly operations for candidate evaluation. In this paper, we propose a new algorithm for computing all the shortest reducts; based on binary cumulative operations over a pair-wise comparison matrix, and a fast candidate evaluation process. Binary cumulative operations save computation time by avoiding repetitive calculations. Furthermore, unlike other algorithms reported in the literature, our candidate evaluation process relays on low-cost operations which reduce the runtime in most cases. Our experiments over synthetic and real-world decision systems show that our proposal is faster than state of the art algorithms in most decision systems. [ABSTRACT FROM AUTHOR]

Most of the current unsupervised feature selection methods are designed to process only numerical datasets. Therefore, in practical problems, where the objects under study are described through both numerical and non-numerical features (mixed datasets), these methods cannot be directly applied. In this work, we propose a new unsupervised filter feature selection method that can be used on datasets with both numerical and non-numerical features. The proposed method is inspired by the spectral feature selection, by using together a kernel and a new spectrum based feature evaluation measure for quantifying the feature relevance. Experiments on synthetic datasets show that in the 99% of the cases where the relevant features are known our method identifies and ranks the most relevant features at the beginning of a sorted list. Additionally, we contrast our method against state-of-the-art unsupervised filter methods over real datasets, and our method in most cases significantly outperforms them. [ABSTRACT FROM AUTHOR]

Abstract: The k nearest neighbor (k-NN) classifier has been a widely used nonparametric technique in Pattern Recognition, because of its simplicity and good performance. In order to decide the class of a new prototype, the k-NN classifier performs an exhaustive comparison between the prototype to classify and the prototypes in the training set T. However, when T is large, the exhaustive comparison is expensive. For this reason, many fast k-NN classifiers have been developed, some of them are based on a tree structure, which is created during a preprocessing phase using the prototypes in T. Then, in a search phase, the tree is traversed to find the nearest neighbor. The speed up is obtained, while the exploration of some parts of the tree is avoided using pruning rules which are usually based on the triangle inequality. However, in soft sciences as Medicine, Geology, Sociology, etc., the prototypes are usually described by numerical and categorical attributes (mixed data), and sometimes the comparison function for computing the similarity between prototypes does not satisfy metric properties. Therefore, in this work an approximate fast k most similar neighbor classifier, for mixed data and similarity functions that do not satisfy metric properties, based on a tree structure (Tree k-MSN) is proposed. Some experiments with synthetic and real data are presented. [Copyright &y& Elsevier]