Yaojin Lin

Feature selection via neighborhood multi-granulation fusion

Feature selection is an important data preprocessing technique, and has been widely studied in data mining, machine learning, and granular computing. However, very little research has considered a multi-granulation perspective. In this paper, we present a new feature selection method that selects distinguishing features by fusing neighborhood multi-granulation. We first use neighborhood rough sets as an effective granular computing tool, and analyze the influence of the granularity of neighborhood information. Then, we obtain all feature rank lists based on the significance of features in different granularities. Finally, we obtain a new feature selection algorithm by fusing all individual feature rank lists. Experimental results show that the proposed method can effectively select a discriminative feature subset, and performs as well as or better than other popular feature selection algorithms in terms of classification performance.

Relations of reduction between covering generalized rough sets and concept lattices

The reduction theory plays an important role in data analysis. This paper studies the relation between the reduction of a covering and the attribute reduction of a concept lattice. The reduction of a covering from the perspective of concept lattices is investigated. Conversely, the attribute reduction of a formal context is studied in the framework of covering generalized rough sets. The results in this paper show that the reduction of a covering can be viewed as the attribute reduction of a derivative formal context. Moreover, every reduct of a given formal context can be seen as the reduct of an induced covering. As an application of the theoretical results, an approach to the attribute reduction of concept lattices based on covering generalized rough sets is proposed. Furthermore, experiments are given to show the effectiveness of the proposed method.

Matrix-based set approximations and reductions in covering decision information systems

In this paper, we propose matrix-based methods for computing set approximations and reducts of a covering decision information system. First, some matrices and matrix operations are introduced to compute the set approximations, and further to compute the positive region of a covering decision system. Second, the notions of minimal and maximal descriptions in a covering decision system are proposed which can be easily obtained by the matrix-based methods. Then the minimal and maximal descriptions are employed to construct a new discernibility matrix. We claim that by using the minimal and maximal descriptions, we can dramatically reduce the total number of discernibility sets that need to be computed in the new discernibility matrix, thus dramatically reducing the computational time for finding all reducts and one optimal reduct of a covering decision system. In the end, several numerical experiments are conducted to examine the efficiency of the proposed methods. We propose matrix-based methods for computing covering set approximations.We propose matrix-based methods for computing covering reducts.A more efficient discernibility matrix is proposed.

Multi-label feature selection based on neighborhood mutual information

Graphical abstractDisplay Omitted HighlightsDifferent from the traditional multi-label feature selection, the proposed algorithm derives from different cognitive viewpoints.A simple and intuitive metric to evaluate the candidate features is proposed.The proposed algorithm is applicable to both categorical and numerical features.Our proposed method outperforms some other state-of-the-art multi-label feature selection methods in our experiments. Multi-label learning deals with data associated with a set of labels simultaneously. Like traditional single-label learning, the high-dimensionality of data is a stumbling block for multi-label learning. In this paper, we first introduce the margin of instance to granulate all instances under different labels, and three different concepts of neighborhood are defined based on different cognitive viewpoints. Based on this, we generalize neighborhood information entropy to fit multi-label learning and propose three new measures of neighborhood mutual information. It is shown that these new measures are a natural extension from single-label learning to multi-label learning. Then, we present an optimization objective function to evaluate the quality of the candidate features, which can be solved by approximating the multi-label neighborhood mutual information. Finally, extensive experiments conducted on publicly available data sets verify the effectiveness of the proposed algorithm by comparing it with state-of-the-art methods.

Feature selection based on quality of information

Feature selection as one of the key problems of data preprocessing is a hot research topic in pattern recognition, machine learning, and data mining. Evaluating the relevance between features based on information theory is a popular and effective method. However, very little research pays attention to the distinguishing ability of feature, i.e., the degree of a feature distinguishes a given sample with other samples. In this paper, we propose a new feature selection method based on the distinguishing ability of feature. First, we define the concept of maximum-nearest-neighbor, and use this concept to discriminate the nearest neighbors of samples. Then, we present a new measure method for evaluating the quality of feature. Finally, the proposed algorithm is tested on benchmark datasets. Experimental results show that the proposed algorithm can effectively select a discriminative feature subset, and performs as well as or better than other popular feature selection algorithms.

The relationship between attribute reducts in rough sets and minimal vertex covers of graphs

The problems to find attribute reduction in rough sets and to obtain the minimal vertex cover for graphs are both NP-hard problems. This paper studies the relationship between the two problems. The vertex cover problem for graphs from the perspective of rough sets is first investigated. The attribute reduction of an information system is then studied in the framework of graph theory. The results in this paper show that finding the minimal vertex cover of a graph is equivalent to finding the attribute reduction of an information system induced from the graph. Conversely, the attribute reduction computation can be translated into the calculation of the minimal vertex cover of a derivative graph. Finally, a new algorithm for the vertex cover problem based on rough sets is presented. Furthermore, experiments are conducted to verify the effectiveness of the proposed method.

Fast approach to knowledge acquisition in covering information systems using matrix operations

Covering rough set theory provides an effective approach to dealing with uncertainty in data analysis. Knowledge acquisition is a main issue in covering rough set theory. However, the original rough set methods are still expensive for this issue in terms of time consumption. To further improvement, we propose fast approaches to knowledge acquisition in covering information systems by employing novel matrix operations. Firstly, several matrix operations are introduced to compute set approximations and reducts of a covering information system. Then, based on the proposed matrix operations, the knowledge acquisition algorithms are designed. In the end, experiments are conducted to illustrate that the new algorithms can dramatically reduce the time consumptions for computing set approximations and reducts of a covering information system, and the larger the scale of a data set is, the better the new algorithms perform.

An effective collaborative filtering algorithm based on user preference clustering

Collaborative filtering is one of widely used recommendation approaches to make recommendation services for users. The core of this approach is to improve capability for finding accurate and reliable neighbors of active users. However, collected data is extremely sparse in the user-item rating matrix, meanwhile many existing similarity measure methods using in collaborative filtering are not much effective, which result in the poor performance. In this paper, a novel effective collaborative filtering algorithm based on user preference clustering is proposed to reduce the impact of the data sparsity. First, user groups are introduced to distinguish users with different preferences. Then, considering the preference of the active user, we obtain the nearest neighbor set from corresponding user group/user groups. Besides, a new similarity measure method is proposed to preferably calculate the similarity between users, which considers user preference in the local and global perspectives, respectively. Finally, experimental results on two benchmark data sets show that the proposed algorithm is effective to improve the performance of recommender systems.

A new nearest neighbor classifier via fusing neighborhood information

The nearest neighbor (NN) classification is a classical and yet effective technique in machine learning and data mining communities. However, its performance depends crucially on the distance function used to compute distance between samples. In this paper, we first define the concept of [emailxa0protected]?s neighborhood and present two related criteria according to neighborhood influence. Then, the influence of [emailxa0protected]?s neighborhood is comprehensively considered when computing the distances between the query and training samples. Finally, we propose an improved nearest neighbor classification algorithm via fusing neighborhood information. The proposed method can precisely characterize the distance among samples as well as enhance the predictive power of classifier to some extent. The experimental results show that the proposed algorithm basically outperforms classical nearest neighbor classifier and some other state-of-the-art classification methods.

A rough set method for the minimum vertex cover problem of graphs

Graphical abstractThe above figure reveals that the new proposed method based on rough sets (VCAR) performs better than the Greedy, List and VSA algorithms on the ratio values for all the graphs listed in the data set. The ratio value is an important index to measure the quality of a solution derived by an algorithm for the minimum vertex cover problem (MVCP), which is defined as Value/Optimum, where Value is the value of a solution found by an algorithm, and Optimum is the optimal solution value. Note that for an approximate algorithm for MVCP, the smaller the ratio value is, the better the algorithm performs. The value found by the VCAR method is quite close to the optimal solution value. We can infer that the new presented algorithm (VCAR) is an effective method for MVCP. Display Omitted HighlightsWe study the NP-hard problem of finding a minimum vertex cover of graphs based on rough sets.The problem of finding a minimum vertex cover of graphs can be translated into the problem of finding an optimal reduct of a decision information table in rough sets.A new method based on rough sets is proposed to compute the minimum vertex cover of a given graph. The minimum vertex cover problem is a classical combinatorial optimization problem. This paper studies this problem based on rough sets. We show that finding the minimal vertex cover of a graph can be translated into finding the attribute reduction of a decision information table. At the same time, finding a minimum vertex cover of graphs is equivalent to finding an optimal reduct of a decision information table. As an application of the theoretical framework, a new algorithm for the minimum vertex cover problem based on rough sets is constructed. Experiments show that the proposed algorithm gets better performance in terms of the ratio value when compared with some other algorithms.