Search

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]

The class imbalance problem is a challenge in supervised classification, since many classifiers are sensitive to class distribution, biasing their prediction towards the majority class. Usually, in imbalanced databases, contrast pattern miners extract a very large collection of patterns from the majority class but only a few patterns (or none) from the minority class. It causes that minority class objects have low support and they could be identified as noise and consequently discarded by the contrast pattern based classifier biasing the results towards the majority class. In the literature, the class imbalance problem is commonly faced by applying resampling methods. Therefore, in this paper, we present a study about the impact of using resampling methods for improving the performance of contrast pattern based classifiers in class imbalance problems. Experimental results using standard imbalanced databases show that there are statistically significant differences between using the classifier before and after applying resampling methods. Moreover, from this study, we provide a guide based on the class imbalance ratio for selecting a resampling method that jointly with a contrast pattern based classifier allows us to have good results in a class imbalance problem. [ABSTRACT FROM AUTHOR]

The representation of hieroglyphs in retrieval systems represents a great challenge since these systems' results highly depend on the used representation. In the literature, the most successful works in this area compute local descriptors from a set of points of interest taken from the entire hieroglyph and use them under the BoVW model to represent hieroglyphs. Unlike these works, this paper presents a way to extract segments from hieroglyphs and, by encoding the extracted segments through local descriptors, proposes to use them under the Bag of Visual Words (BoVW) model to represent hieroglyphs. Our experiments show that the proposed representation allows us to obtain retrieval results that overcome those reported by using state of the art representations. • Hieroglyph representation for retrieval based on encoding hieroglyph segments. • Good quality hieroglyph retrieval following the Bag of Visual Words Model. • Automatic decomposition of hieroglyphs in segments. [ABSTRACT FROM AUTHOR]

Most understandable classifiers are based on contrast patterns, which can be accurately mined from decision trees. Nevertheless, tree diversity must be ensured to mine a representative pattern collection. In this paper, we performed an experimental comparison among different diversity generation procedures. We compare diversity generated by each procedure based on the amount of total, unique, and minimal patterns extracted from the induced tree for different minimal support thresholds. This comparison, together with an accuracy and abstention experiment, shows that Random Forest and Bagging generate the most diverse and accurate pattern collection. Additionally, we study the influence of data type in the results, finding that Random Forest is best for categorical data and Bagging for numerical data. Comparison includes most known diversity generation procedures and three new deterministic procedures introduced here. These deterministic procedures outperform existing deterministic method, but are still outperformed by random procedures. [ABSTRACT FROM AUTHOR]

Abstract: Most of the current algorithms for mining association rules assume that two object subdescriptions are similar when they are exactly equal, but in many real world problems some other similarity functions are used. Commonly these algorithms are divided in two steps: Frequent pattern mining and generation of interesting association rules from frequent patterns. In this work, two algorithms for mining frequent similar patterns using similarity functions different from the equality are proposed. Additionally, the GenRules Algorithm is adapted to generate interesting association rules from frequent similar patterns. Experimental results show that our algorithms are more effective and obtain better quality patterns than the existing ones. [Copyright &y& Elsevier]

Abstract: Most clustering algorithms organize a collection of objects into a set of disjoint clusters. Although this approach has been successfully applied in unsupervised learning, there are several applications where objects could belong to more than one cluster. Overlapping clustering is an alternative in those contexts like social network analysis, information retrieval and bioinformatics, among other problems where non-disjoint clusters appear. In addition, there are environments where the collection changes frequently and the clustering must be updated; however, most of the existing overlapping clustering algorithms are not able to efficiently update the clustering. In this paper, we introduce a new overlapping clustering algorithm, called DClustR, which is based on the graph theory approach and it introduces a new strategy for building more accurate overlapping clusters than those built by state-of-the-art algorithms. Moreover, our algorithm introduces a new strategy for efficiently updating the clustering when the collection changes. The experimentation conducted over several standard collections shows the good performance of the proposed algorithm, wrt. accuracy and efficiency. [Copyright &y& Elsevier]

Abstract: In this paper, we introduce an efficient algorithm for mining discriminative regularities on databases with mixed and incomplete data. Unlike previous methods, our algorithm does not apply an a priori discretization on numerical features; it extracts regularities from a set of diverse decision trees, induced with a special procedure. Experimental results show that a classifier based on the regularities obtained by our algorithm attains higher classification accuracy, using fewer discriminative regularities than those obtained by previous pattern-based classifiers. Additionally, we show that our classifier is competitive with traditional and state-of-the-art classifiers. [Copyright &y& Elsevier]

Contrast patterns, which lie in the core of most understandable classifiers, are frequently evaluated by quality measures. Since many different quality measures are available, they should be compared to select the most appropriate for each applications. This paper introduces a method to compare quality measures, using a set of mined patterns and a collection of objects not used for mining. The comparison is performed by correlating quality values with a quality estimation of the patterns. Additionally, a meta-learning study is performed to show that combining quality measures could be better than using the best single measures in isolation. The results of this paper can help researchers to create new quality measures or to find new combinations of quality measures to create better understandable classification systems. [ABSTRACT FROM AUTHOR]

Contrast pattern-based classifiers are an important family of both understandable and accurate classifiers. Nevertheless, these classifiers do not achieve good performance on class imbalance problems. In this paper, we introduce a new contrast pattern-based classifier for class imbalance problems. Our proposal for solving the class imbalance problem combines the support of the patterns with the class imbalance level at the classification stage of the classifier. From our experimental results, using highly imbalanced databases, we can conclude that our proposed classifier significantly outperforms the current contrast pattern-based classifiers designed for class imbalance problems. Additionally, we show that our classifier significantly outperforms other state-of-the-art classifiers not directly based on contrast patterns, which are also designed to deal with class imbalance problems. [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: Association rule mining and classification are important tasks in data mining. Using association rules has proved to be a good approach for classification. In this paper, we propose an accurate classifier based on class association rules (CARs), called CAR-IC, which introduces a new pruning strategy for mining CARs, which allows building specific rules with high confidence. Moreover, we propose and prove three propositions that support the use of a confidence threshold for computing rules that avoids ambiguity at the classification stage. This paper also presents a new way for ordering the set of CARs based on rule size and confidence. Finally, we define a new coverage strategy, which reduces the number of non-covered unseen-transactions during the classification stage. Results over several datasets show that CAR-IC beats the best classifiers based on CARs reported in the literature. [Copyright &y& Elsevier]

• A systematic evaluation of filter Unsupervised Feature Selection methods is presented. • The most popular and recent filter UFS methods are included in our study. • The evaluation of the filter UFS methods followed the standards in the literature. • A general discussion based on the results of the evaluated methods is provided. • Some guidelines for the use of the evaluated filter UFS methods is also provided. Unsupervised Feature Selection (UFS) has aroused great interest in the last years because of its practical significance and application on a large variety of problems in expert and intelligent systems where unlabeled data appear. Specifically, Unsupervised Feature Selection methods based on the filter approach have received more attention due to their efficiency, scalability, and simplicity. However, in the literature, there are no comprehensive studies for assessing such UFS methods when they are applied, under the same conditions, to a wide variety of real-world data. To fill this gap, in this paper, we present a comprehensive empirical and systematic evaluation of the most popular and recent filter UFS methods, evaluating their performance in terms of clustering, classification, and runtime. The filter methods used in our study were applied on 50 datasets from the UCI Machine Learning Repository and 25 high dimensional datasets from the ASU Feature Selection Repository. To evaluate if the outcomes obtained by the assessed methods are statistically significant, the Friedman test and Holm post hoc procedure were applied in the clustering and classification results. From our experiments, we provide some practical guidelines and insights for the use of the filter UFS methods analyzed in our study. [ABSTRACT FROM AUTHOR]

Highlights • Outlier detection is important for improving clustering results. • Over detection of outliers leads to information loss. • Our proposal reduces the number of over-detected outliers. • Experiments show that clustering quality can be improved, while runtime is reduced. Abstract Expert and intelligent systems use a variety of machine learning techniques to obtain and understand the information inherent in the data. Clustering is one of these techniques, which has become important and popular since it allows classifying an unlabeled dataset into clusters of similar objects. There are many clustering algorithms that have been proposed in the literature. From these algorithms, the Cross-Clustering algorithm is one of the most recent clustering algorithms for partial clustering (clustering where not necessarily all the objects are grouped into clusters), which has provided good results allowing estimating a suitable set of clusters, as well as eliminating outliers. However, this algorithm tends to eliminate too many objects as outliers, which leads to discard a lot of non-outlier objects. Additionally, the Cross-Clustering algorithms spends a lot of time evaluating several combinations of clusterings, trying to determine a suitable number of clusters. To overcome these problems, in this paper, an improved version of the Cross-Clustering algorithm (ICC) is proposed. ICC changes the clustering algorithm used for detecting outliers, as well as it modifies the way outliers are detected. Moreover, a stop criterion allowing to make a fast decision on the estimation of a suitable number of cluster, is also introduced. The performance of the improved Cross-Clustering algorithm is compared with the original algorithm on artificial and real datasets. Our results show that ICC improves the original algorithm and other state of the art clustering algorithms; in both, runtime and clustering quality. [ABSTRACT FROM AUTHOR]

Frequent pattern mining is considered a key task to discover useful information. Despite the quality of solutions given by frequent pattern mining algorithms, most of them face the challenge of how to reduce the number of frequent patterns without information loss. Frequent itemset mining addresses this problem by discovering a reduced set of frequent itemsets, named closed frequent itemsets , from which the entire frequent pattern set can be recovered. However, for frequent similar pattern mining , where the number of patterns is even larger than for Frequent itemset mining, this problem has not been addressed yet. In this paper, we introduce the concept of closed frequent similar pattern mining to discover a reduced set of frequent similar patterns without information loss. Additionally, a novel closed frequent similar pattern mining algorithm, named CFSP-Miner , is proposed. The algorithm discovers frequent patterns by traversing a tree that contains all the closed frequent similar patterns. To do this efficiently, several lemmas to prune the search space are introduced and proven. The results show that CFSP-Miner is more efficient than the state-of-the-art frequent similar pattern mining algorithms, except in cases where the number of frequent similar patterns and closed frequent similar patterns are almost equal. However, CFSP-Miner is able to find the closed similar patterns, yielding a reduced size of the discovered frequent similar pattern set without information loss. Also, CFSP-Miner shows good scalability while maintaining an acceptable runtime performance. [ABSTRACT FROM AUTHOR]

Abstract: Water quality assessment is an important activity for controlling harmful crisis in aquaculture systems. The objective of our study was to develop a new Water Quality Index focused on monitoring of shrimp farms; detecting poor water quality and preventing negative effects in the ecosystem. Usually, several water quality parameters are monitored and measured in a shrimp farm during a farming period. Those parameters are classified according to their negative effects in the ecosystem and their respective allowed limits are also defined. The proposed Water Quality Index assigns a priority level to each water parameter through a new analytical hierarchical process (AHP), which allows an accurate assessment of the water quality. Our proposed index was applied to assess the water quality condition in extensive shrimp farms in Mexico. A comparison between our approach and those proposed in the literature shows its good performance when real environments are assessed. [Copyright &y& Elsevier]

Abstract: In recent years, artificial intelligence methods have been used for the treatment of environmental problems. This work, presents two models for assessment and prediction of air quality. First, we develop a new computational model for air quality assessment in order to evaluate toxic compounds that can harm sensitive people in urban areas, affecting their normal activities. In this model we propose to use a Sigma operator to statistically asses air quality parameters using their historical data information and determining their negative impact in air quality based on toxicity limits, frequency average and deviations of toxicological tests. We also introduce a fuzzy inference system to perform parameter classification using a reasoning process and integrating them in an air quality index describing the pollution levels in five stages: excellent, good, regular, bad and danger, respectively. The second model proposed in this work predicts air quality concentrations using an autoregressive model, providing a predicted air quality index based on the fuzzy inference system previously developed. Using data from Mexico City Atmospheric Monitoring System, we perform a comparison among air quality indices developed for environmental agencies and similar models. Our results show that our models are an appropriate tool for assessing site pollution and for providing guidance to improve contingency actions in urban areas. [Copyright &y& Elsevier]

Abstract: The continuous monitoring of physical, chemical and biological parameters in shrimp culture is an important activity for detecting potential crisis that can be harmful for the organisms. Water quality can be assessed through toxicological tests evaluated directly from water quality parameters involved in the ecosystem; these tests provide an indicator about the water quality. The aim of this study is to develop a fuzzy inference system based on a reasoning process, which involves aquaculture criteria established by official organizations and researchers for assessing water quality by analyzing the main factors that affect a shrimp ecosystem. We propose to organize the water quality parameters in groups according to their importance; these groups are defined as daily, weekly and by request monitoring. Additionally, we introduce an analytic hierarchy process to define priorities for more critical water quality parameters and groups. The proposed system analyzes the most important parameters in shrimp culture, detects potential negative situations and provides a new water quality index (WQI), which describes the general status of the water quality as excellent, good, regular and poor. The Canadian water quality and other well-known hydrological indices are used to compare the water quality parameters of the shrimp water farm. Results show that WQI index has a better performance than other indices giving a more accurate assessment because the proposed fuzzy inference system integrates all environmental behaviors giving as result a complete score. This fuzzy inference system emerges as an appropriated tool for assessing site performance, providing assistance to improve production through contingency actions in polluted ponds. [Copyright &y& Elsevier]

• Fast oversampling method in the string space based on the edit distance. • Good quality oversampling not based on searching nearest neighbors. • Hierarchical synthetic object generation for class imbalance problems. In syntactic and structural pattern recognition, data represented as strings appear in several supervised classification applications. In some situations, data collections show imbalanced class distributions, which typically results in the classifier biasing its performance to the class representing the majority of objects. To solve this problem, some oversampling methods have been proposed for data represented as strings. However, this type of method has been little studied in the literature. Therefore, in this paper, we present an oversampling method for working in string space that balances the minority class and gets better classification results than state-of-the-art oversampling methods, especially for highly imbalanced problems. Furthermore, according to our experiments, the proposed method is much faster than those reported in the literature. [ABSTRACT FROM AUTHOR]