Rough set theory constitutes a sound basis for KDD (Knowledge Discovery and Data Mining) to deal with real world problems systematically. In the paper, we investigate several rough sets based hybrid systems that can be used in a multi-phase KDD process to discover patterns hidden in data in many aspects. We also outline the latest researches and give future directions.

View full abstract

In this paper, we develop a model for forecasting the amount of carbon dioxide emission due to the traffic of urban commuters and students. The model consists of three sub-models: a sub-model for estimating the number of commuters and students, commuter modal choice sub-model with fuzzy reasoning and a sub-model for estimating the amount of carbon dioxide emission. Using this model, we analyze and predict the effect of policies to shift the commuters' travel mode from private car to public transportation for decreasing amount of carbon dioxide emission.

View full abstract

The machine learning method is proposed to learn techniques of specialists. A machine has to learn techniques by trial and error when there are not enough training data. Reinforcement learning is a powerful machine learning system, which is able to learn without giving training data to a learning unit. But it is impossible for the reinfocement learning to support large environments because the number of if-then rules, which explain a relationship between one environment and one action, is a huge combination. We propose a new reinforcement learning with fuzzy environment evaluation. The fuzzy environment evaluation rule shows a relationship between one environment and one evaluation. This machine learning system is made up from a fuzzy evaluation, an environment simulator and MinMax search. The learning unit renews the evaluation in every action. The fuzzy evaluation of inexperienced environments is reasoned by fuzzy rules. The fuzzy evaluation, the environment simulator and Min Max search present the best policy in a huge environment. We dealt with chess as an example of target environment. Then we will show the excellent results of chess against the GNU chess. The fuzzy evaluation of inexperienced environments is reasoned by fuzzy rule sets, machine has.

View full abstract

There are many researches about motion planning problems. In this field, main researches are to generate motion for specific tasks or robots without previously generated motions. We research a motion planning having a mechanism of reusing acquired motions. We introduce a hierarchical knowledge to realize a mechanism of reusing acquired motions. In this paper, we adapt tree-based representation for expressing motions and adopt genetic programming as learning method. We construct the motion planning system using the hierarchical knowledge. We apply the proposed method to a six-legged locomotion robot and show its availability.

View full abstract

The purpose of this study is to develop a new evacuation simulation model by artificial life approach and fuzzy inference. This model regards evacuees as agents of artificial life which are given some simple rules, and deals with agents' psychology such as an awareness of danger by fuzzy inference. There are three types of agents: ordinary people, guides and old people. Only guides can communicate the information of escape routes initially. However, agents whose values of awareness of danger become high to a certain degree can communicate the information of escape routes from other agents which know escape routes. These simple rules make agents behave the group behavior such as the detouring route. The proposed model uses this group behavior as the evacuation behavior. The proposed model has the following features: (1) it can handle more precise and more complex evacuation behavior than previous studies by means of the interaction of agents which is information transmission and psychology between agents; (2) it can get some knowledge about evacuation behavior.

View full abstract

We present a detection method of influential observations for possibilistic linear regression analysis, where input data and output data are represented by L-R fuzzy numbers. In this method, we use a sensitivity analysis using concepts of duality and its allowable right-hand side range in linear programming, in order to evaluate an effect of influential observations on fuzzy regression coefficients. The proposed method deals with two constraints corresponding to one observation using the dual variable and their allowable right-hand side ranges in primal problem. We demonstrate an application of the proposed method on consumer decision research that examines effects of multiattributes on behavioral intention.

View full abstract

We have two important assumptions, the perfect imformation and the rational behavior, when we evaluate the infrastructure constructing by the ecobomic theory. But we think there is always fizziness in those behaviors. In this paper, we built the benefit evaluation model by using the fuzzy utility, in order to consider the fuzziness of decision-making, We introduced the fuzziness for the utility maximizing behavior in the general equilibrium model, and showed the possibility to measure the benefit as fuzzy. We carried out the simulation analysis for the assumption case. Though we set the 1% fuzziness, the benefit had the large width. We must plan the policy as being care for its facts.

View full abstract

Residue Fluid Catalytic Cracking Unit (RFCCU) is the heart of the modern petroleum refinery. There has been a strong demand for decreasing the electric power and manpower in operating RFCCU. A combined control system has been constructed for satisfying this demand by controlling the reactor/regenerator differential pressure of RFCCU. The problem of this control system is that the analyzer of oxygen component, which is one of the important input variables of this control system, lacks reliability. A modeling method using neural networks and a linear regressive model has been developed. But the accuracy of this predictor became unsatisfactory in the case of drastic changes of operating conditions, such as feed oil change. This paper presents a fuzzy partition method based on mutual entropy for identification of prediction model of oxygen component. This method uses ID3 algorithm to select predictor variables. The input space is fuzzily divided based on the mutual entropy. A linear regressive equation is identified in each sub-space, and a fuzzy model is constructed. The obtained predictor shows a satisfactory performance even in the case where the operating conditions are changed drastically. Continuous operation of the combined controller is made possible with this predictor. The differential pressure is stabilized, and a considerable amount of electric power is reduced. The implementation of this control system can also reduce the manpower needed for the operation of RFCCU.

View full abstract

Fuzzy c-Varieties (FCV) clustering proposed by Bezdek et al. is a linear clustering method whose prototypes are linear carieties and can be regarded as a technique for extracting local principal components. In spite of its usefulness, the FCV algorithm cannot deal with an incomplete data set including missing values without elimination or imputation of data. In this paper, we propose a method for partitioning an incomplete data set including missing values into several fuzzy clusters using local principal components. First, FCV clustering is defined as the technique for the extraction of local principal components based on the minimization of the least square criterion, which performs the lower rank approximation of the data matrix. While the objective function of FCV clustering is based on the minimization of the distances between data points and prototypical linear varieties, the same objective function can be derived from the least square criterion under a certain condition. Second, a new technique for dealing with incomplete data sets is proposed by extending the method to extract local principal components. Numerical example shows the characteristic properties of our method.

View full abstract

The aim of the present paper is to discuss L_1 based fuzzy c-means for fazzy data. Data unit is supposed to be Cartesian product of triangular type fuzzy numbers. The norm between a fuzzy data and a cluster center is defined using Maximum and Minimum Distances. Fuzzy c-means algorithm is an alternate optimization procedure of the cluster centers and the memberships, where the solution of cluster centers for fuzzzy data can not be obtained directly. The algorithms based on L_1 metric for the solution of cluster centers are developed in this paper. Using these algorithms, exact alternate optimization procedure is obtained. Numerical examples show that the results for the data with uncertainties are different from the results for the data without uncertainties.

View full abstract

For decision making problems involving uncetainty, there exist two typical approaches: a stochastic programming aiming to optimize based on the probability theory and a fuzzy programming using fuzzy concepts to deal with the ambiguity in the decision making situation, and they have been developed in various ways. In this paper, as a hybrid of the stochastic programming and the fuzzy programming, we focus on multiobjective linear programming problems with random variables in the right-hand side of constraints. After reformulating them based on a simple recourse model to introduce penalty with respect to the difference between the left-hand side and the right-hand side of constraints, we attempt to apply the interactive fuzzy satisficing method to the formulated multiobjective simple recourse problems.

View full abstract

This paper proposes a new computational method in possibilistic clustering and compares solutions of them with those by the fuzzy c-means using probabilistic partitions. Two objective functions for both the probabilistic partitions in fuzzy c-means and possibilistic partitions in possibilistic clustering are considered for this purpose, namely, a regularized objective function obtained from the standard fuzzy c-means and that of the entropy regularization are employed. Relations between solutions for probabilistic and possibilistic partitions are investigated. Ordinary algorithm in possibilistic clustering is shown to be improved by using many initial cluster centers instead of the c centers, whereby the number of clusters is estimated after the iteration of the new algorithm. Classification functions using this method is moreover proposed. Numerical results using the iris data show effectiveness of the present method of computation.

View full abstract