Jiucheng Xu

An ensemble feature selection technique for cancer recognition

Correlation-based feature selection (CFS) using neighborhood mutual information (NMI) and particle swarm optimization (PSO) are combined into an ensemble technique in this paper. Based on this observation, an efficient gene selection algorithm, denoted by NMICFS-PSO, is proposed. Several cancer recognition tasks are gathered for testing the proposed technique. Moreover, support vector machine (SVM), integrated with leave-one-out cross-validation and served as a classifier, is employed for six classification profiles to calculate the classification accuracy. Experimental results show that the proposed method can reduce the redundant features effectively and achieve superior performance. The classification accuracy obtained by our method is higher in five out of the six gene expression problems as compared with that of other classifi cation methods.

New reduction algorithm based on decision power of decision table

The current reduction algorithms based on rough sets still have some disadvantages. First, we indicated their limitations for reduct generation. We modified the mean decision power, and proposed to use the algebraic definition of decision power. To select optimal attribute reduction, the judgment criterion of decision with inequality was presented and some important conclusions were obtained. A complete algorithm for the attribute reduction was designed. Finally, through analyzing the given example, it was shown that the proposed heuristic information was better and more efficient than the others, and the presented in the paper method reduces time complexity and improves the performance. We report experimental results with several data sets from UCI repository and we compare the results with some other methods. The results prove that the proposed method is promising.

Methods and Practices of Three-Way Decisions for Complex Problem Solving

A theory of three-way decisions is formulated based on the notions of three regions and associated actions for processing the three regions. Three-way decisions play a key role in everyday decision-making and have been widely used in many fields and disciplines. A group of Chinese researchers further investigated the theory of three-way decision and applied it in different domains. Their research results are highlighted in an edited Chinese book entitled “Three-way Decisions: Methods and Practices for Complex Problem Solving.” Based on the contributed chapters of the edited book, this paper introduces and reviews most recent studies on three-way decisions.

Knowledge reduction based on granular computing from decision information systems

Efficient knowledge reduction in large inconsistent decision information systems is a challenging problem. Moreover, existing approaches have still their own limitations. To address these problems, in this article, by applying the technique of granular computing, provided some rigorous and detailed proofs, and discussed the relationship between granular reduct introduced and knowledge reduction based on positive region related to simplicity decision information systems. By using radix sorting and hash methods, the object granules as basic processing elements were employed to investigate knowledge reduction. The proposed method can be applied to both consistent and inconsistent decision information systems.

Decision Degree-based Decision Tree Technology for Rule Extraction

Traditional rough set-based approaches to reduct have difficulties in constructing optimal decision tree, such as empty branches and over-fitting, selected attribute with more values, and increased expense of computational effort. It is necessary to investigate fast and effective search algorithms. In this paper, to address this issue, the limitations of current knowledge reduction for evaluating decision ability are analyzed deeply. A new uncertainty measure, called decision degree, is introduced. Then, the attribute selection standard of classical heuristic algorithm is modified, and the new improved significance measure of attribute is proposed. A heuristic algorithm for rule extraction from decision tree is designed. The advantages of this method for rule extraction are that it needn’t compute relative attribute reduction of decision tables, the computation is direct and efficient, and the time complexity is much lower than that of some existing algorithms. Finally, the experiment and comparison show that the algorithm provides more precise and simplified decision rules. So, the work of this paper will be very helpful for enlarging the application areas of rough set theory.

Research of Knowledge Reduction Based on New Conditional Entropy

Although knowledge reduction for a decision table based on discernibility function can be used widely in data classification, there are also many disadvantages needed discussing detailedly on knowledge acquisition. To make some improvement for them, firstly, the concept of a decision table simplified was put forward for removing redundant data. Then based on knowledge granulation and conditional information entropy, the definition of a new conditional entropy, which could reflect the change of decision ability objectively and equivalently and present the concepts and operations in an inconsistent decision table simplified, was given by separating the consistent objects from the inconsistent objects. Furthermore, many propositions and properties for reduction with an inequality were proposed, and a complete knowledge reduction method was implemented. Finally, the experimental results with UCI data sets show that the proposed method of knowledge reduction is an effective technique to deal with complex data sets, and can simplify the structure and improve the efficiency of data classification.

A Greedy Correlation Measure Based Attribute Clustering Algorithm for Gene Selection

This paper proposes an attribute clustering algorithm for grouping attributes into clusters so as to obtain meaningful modes from microarray data. First the problem of attribute clustering is analyzed and neighborhood mutual information is introduced to solve it. Furthermore, an attribute clustering algorithm is presented for grouping attributes into clusters through optimizing a criterion function which is derived from an information measure that reflects the correlation between attributes. Then, by applying this method to gene expression data, meaningful clusters are discovered which assists to capture aspects of gene association patterns. Thus, significant genes containing useful information for gene classification and identification are selected. In the following, the proposed algorithm is employed to six gene expression data sets and a comparison is made with several well known gene selection methods. Experiments show that the greedy correlation measure based attribute clustering algorithm, noted as GCMACA, is more capable of discovering meaningful clusters of genes. Through selecting a subset of genes which have a high significant multiple correlation value with others within clusters, informative genes can be acquired and gene expression of different categories can be identified as well.

A color image segmentation algorithm by integrating watershed with region merging

In order to improve the effectiveness of color image segmentation, a color image segmentation algorithm by integrating watershed with region merging is proposed in this paper. First, the image input is divided into many regions by watershed algorithm, and the phenomenon of the image over segmentation emerges for the details and noise of the image information. Second, the integrated regional distance, which is integrated with the factors, such as the image color information, edge strength and adjacency information, is defined. Third, an algorithm of diminutive region merging is designed to remove the diminutive region. As a result, the color image segmentation is more efficiently realized. Finally, a simulation experiment is implemented with the algorithm proposed, then the analysis is also given in this paper, and it is proved that the algorithm proposed is more effective in the color image segmentation and solves the problem of the over segmentation generated by the watershed algorithm segmentation.