Chen-Xia Jin

A generalized fuzzy ID3 algorithm using generalized information entropy

A fuzzy decision tree is an important tool for knowledge acquisition in uncertain environments. Most of the existing fuzzy decision tree algorithms do not systematically consider the impact of the non-linear characteristics of the membership degree of fuzzy sets; they are therefore unable to integrate uncertainty processing preferences into the selection of extended attributes. This paper initially offers a generalized Hartley metric model and calculation method. We then introduce a fuzzy consciousness function and further provide generalized fuzzy partition entropy for the attribute-selecting heuristic of a fuzzy decision tree. We subsequently propose a generalized fuzzy partition entropy-based fuzzy ID3 algorithm (abbreviated as GFID3) that can support decision making and analyze the performance of the GFID3 through several case-based examples. The experimental results show that the GFID3 algorithm demonstrates better structural characteristics and operability in practical applications and has high computational precision. It ameliorates the deficiencies of existing fuzzy decision tree algorithms and can be used in fields such as complex systems optimization, data mining and intelligent systems.

Feature selection with partition differentiation entropy for large-scale data sets

Feature selection, especially for large data sets, is a challenging problem in areas such as pattern recognition, machine learning and data mining. With the development of data collection and storage technologies, the data has become bigger than ever, thus making it difficult for learning from large data sets with traditional methods. In this paper, we introduce the partition differentiation entropy from the viewpoint of partition in rough sets to measure the significance and uncertainty of attributes, and present a feature selection method for large-scale data sets based on the information-theoretical measurement of attribute significance. Given a large-scale decision information system, the proposed method first divides it into small sub information systems according to the decision classes. Then by computing partition differentiation entropy in the sub-systems, the partition differentiation entropy of the attribute subset in the original decision information system is obtained. Accordingly, the important features are selected based on the value of partition differentiation entropy. The experimental results show that the idea of the proposed method is feasible and valid.

Selection method of the 3PL supplier based on interval satisfaction

Outsourcing is an increasingly important issue pursued by corporations seeking improved efficiency. Logistics outsourcing or third-party logistics (3PL) involves the use of external companies to perform some or all of the firms logistics activities. In this paper, by analyzing the feature and role of third-party logistics, for 3PL provider selection, we put forward basic principle of data integration, establish comprehensive evaluation model of 3PL supplier based on interval satisfaction, and give a kind of interval data weighted fuzzification method; furthermore, starting from the structural feature of fuzzy information, we propose the compound quantification description mode based on centralized quantification value, and establish a fuzzy information comparison method based on synthesizing effect and 3PL supplier selection model, and also analyze the feature through real case. All the results indicate that the 3PL supplier selection model proposed can effectively merge decision preference into decision process and has better operability.

Knowledge change rate-based attribute importance measure and its performance analysis

Attribute importance measure is important in such approaches as data system reduction and, multi-attribute decisions. In this paper, we present knowledge change rate-based attribute importance measures with structural features of fuzzy measure, abbreviated as BCKCRAIM. We discuss theoretical construction strategies and structural features followed by remarks on constructing BCKCRAIM. Finally, experimental results for several examples and UCI data sets show the connections and differences between BCKCRAIM and other attribute importance measures. The advantage of our measure is that it uses attributes set changes to describe knowledge change and associated features between lower and upper approximations of decision classes and knowledge to reflect attribute importance. Our measure can improve feasibility and interpretability; therefore, BCKCRAIM has wide application in such approaches as attributes reduction, feature extraction, information fusion, and expert systems.

A new effect-based roughness measure for attribute reduction in information system

Roughness measure, a quantitative index for processing uncertain information using fuzzy set theory, is the basis of resource management, system optimization and many other decision-making problems. The construction of a roughness measure which reflects different decision preferences has important theoretical and practical value. In this paper, we first analyze the characteristics and shortcomings of the existing methods for measuring roughness. Following this, a new effect-based roughness measure model is established using a combination of a basic measure factor, the lower (or upper) accuracy of rough sets, entitled the effective rough degree (ERD). Next, the characteristics of ERD are analyzed in combination with different synthesis functions, and several further necessary and sufficient conditions are given. Finally, we propose an ERD-based attribute reduction method (abbreviated as ERD-RM), and then discuss the differences and relationships between ERD-RM and existing reduction methods. The theoretical analysis and practical applications show that ERD has good structural features and interpretability and can integrate decision preference into the measure system in a straightforward manner.

Fuzzy optimization model based on synthesizing effect and inequity degree

In this paper, based on the structure of fuzzy information and the mechanism of fuzzy optimization, we propose the concept of quasi-linear fuzzy number; by distinguishing principal index and secondary indices, we give the comparison method based on synthesizing effect combining with the compound qualification strategy of fuzzy information; starting from the essence of constraints, we give a fuzzy optimization model based on synthesizing effect and inequity degree (BID & SE-FOM), and propose an instructive fuzzy genetic algorithm based on principal operation and quasi-linear fuzzy numbers(PO QL-FGA); Finally, we analyze the performance of PO QL-FGA by using Markov chain theory, and further explain the application of quasi-linear fuzzy numbers by a concrete example.

Decision Tree Inductive Learning Algorithm Based on Removing Noise Gradually

When noise exists in case base, high quality knowledge is hard to obtain by ID3 algorithm. For the weakness, by introducing the concept of second learning, the noisy data can be removed, which not only develop the decision tree, but also it can make good structure tree generate. So that we can abstract good rules information, and make the desirable tree more accurate. Especially, the more the data can be mined by decision tree algorithm, the better the efficiency and performance of the algorithm is, and the more obvious the superiority of algorithm is. This paper states the basic idea of algorithm, implementation process, performance analysis and accuracy proof in detail.

A new compound arithmetic crossover-based genetic algorithm for constrained optimisation in enterprise systems

ABSTRACT In many real industrial applications, the integration of raw data with a methodology can support economically sound decision-making. Furthermore, most of these tasks involve complex optimisation problems. Seeking better solutions is critical. As an intelligent search optimisation algorithm, genetic algorithm (GA) is an important technique for complex system optimisation, but it has internal drawbacks such as low computation efficiency and prematurity. Improving the performance of GA is a vital topic in academic and applications research. In this paper, a new real-coded crossover operator, called compound arithmetic crossover operator (CAC), is proposed. CAC is used in conjunction with a uniform mutation operator to define a new genetic algorithm CAC10-GA. This GA is compared with an existing genetic algorithm (AC10-GA) that comprises an arithmetic crossover operator and a uniform mutation operator. To judge the performance of CAC10-GA, two kinds of analysis are performed. First the analysis of the convergence of CAC10-GA is performed by the Markov chain theory; second, a pair-wise comparison is carried out between CAC10-GA and AC10-GA through two test problems available in the global optimisation literature. The overall comparative study shows that the CAC performs quite well and the CAC10-GA defined outperforms the AC10-GA.

Stability analysis and fuzzy control design for near space vehicle re-entry attitude dynamics

This paper investigates a novel composite control scheme of fuzzy control, which is used for near space vehicle (NSV) attitude dynamics during the re-entry phase. For the complicated flight condition during re-entry phase, the Takagi and Sugeno (T-S) fuzzy model is employed to approximate near space vehicle attitude dynamics, on the basis of the linear matrix inequality (LMI) technique and parametric optimization method, the composite fuzzy control algorithms are derived. Unlike earlier studies of fuzzy control systems on an LMI framework, this study develops a supervisory control approach, such that a fuzzy controller can be synthesized more efficiently. Finally, simulation result demonstrates the effectiveness and potential of the proposed technique.

A rule extraction algorithm based on attribute importance

Classification algorithm is a kind of important technology in data mining, and the most commonly used is decision tree learning. In the process of constructing a decision tree, the selecting criteria of splitting attributes will directly affect the classification results. And the attribute selection of the traditional decision tree algorithm is based on information theory. In this paper, by combining with rough sets theory, we propose a new rules extraction algorithm based on attributes importance and dependence. Compared with the other algorithm, our algorithm is simple, by which we can obtain comprehensive rules without redundancy, and it also gives rule mining process with higher reliability.