Kyung-Whan Oh

A validity measure for fuzzy clustering and its use in selecting optimal number of clusters

Cluster analysis has been playing an important role in solving many problems in pattern recognition and image processing. If fuzzy cluster analysis is to make a significant contribution to engineering applications, much more attention must be paid to fundamental decision on the number of clusters in data. It is related to cluster validity problem of how well it has identified the structure that is present in the data. In this paper, we define I/sub G/ as a fuzzy clustering validity function which measures the overall average compactness and separation of fuzzy c-partition and propose a new approach to selecting optimal number of clusters using the measurement value of I/sub G/. This approach uses relative values and normalized value of I/sub G/ and it does not require human interpretation. It is compared with conventional validity functions, partition coefficient and CSC index, on the several data sets.

ASA and its application to multi-criteria decision making (case study)

Aggregation plays an important role in many applications related to the development of intelligent systems. In a fuzzy environment the existing aggregation operators are generally the t-norm, t-conorm, mean operators, Yagers operator and γ-operator. However, these aggregation operators do not properly reflect the situation in the aggregation process. That is, these types of aggregation operators are independent of the aggregation situation. In order to solve these problems we propose a new aggregation method to reflect the situation in the aggregation process. It is the aggregation based on situation assessment (ASA) method. It consists of the situation assessment model (SAM) and the ASA algorithm. In this ASA method, the SAM is utilized to reflect the situation in the aggregation process. This model generates the parameter, which is controlled by decision-maker. It indicates the current degree of aggregation situation. Therefore, our method can be adapted to certain situations by using the parameter. In this paper, ASA method is applied to the multi-criteria decision making environment.

A fuzzy reinforcement function for the intelligent agent to process vague goals

The intelligent agent is one of the most interesting fields of Artificial Intelligence study. Generally, very many kinds of the intelligent agent receive the users goal and they try to solve it with their expert knowledge. The users goals or requests can be represented with the human language, and they contain the uncertainties of the human knowledge. While the intelligent agent must represent these vague goals and understand the users desires or intentions, there have not been enough researches done for the intelligent agents to express the users goals. In this paper, we propose a new method to represent the vague goals as well as the uncertain environments. We suggest a fuzzy goal and a fuzzy state representation. We extend the traditional reinforcement learning to the fuzzy reinforcement learning with defining the fuzzy reinforcement function by using the fuzzy goal and the fuzzy stare. We, also propose a new Fuzzy Q-Learning algorithm. The experiment results show the better performance of the learning, and the reasonability of the fuzzy reinforcement learning.

A face recognition system using fuzzy logic and artificial neural network

The authors have developed a method to extract a feature vector that is very important to recognizing facial images. The eye blinking method was used to get the location of the eyes roughly. Then a feature vector was obtained using locations and distances between feature points, that is the eyes, the nose, the mouth and the outline of the face. To make the feature vector invariant to the size of the facial image, it was normalized. Fuzzy linguistic variables were used instead of real numbers to represent the approximate distance between feature points. These fuzzified feature vectors were learned by an artificial neural network and used to recognize a facial image in the recognition phase. The face recognizer could recognize all learned persons correctly in spite of variations.<<ETX>>

Prediction of Molecular Bioactivity for Drug Design Using a Decision Tree Algorithm

A machine learning-based approach to the prediction of molecular bioactivity in new drugs is proposed. Two important aspects are considered for the task: feature subset selection and cost-sensitive classification. These are to cope with the huge number of features and unbalanced samples in a dataset of drug candidates. We designed a pattern classifier with such capabilities based on information theory and re-sampling techniques. Experimental results demonstrate the feasibility of the proposed approach. In particular, the classification accuracy of our approach was higher than that of the winner of KDD Cup 2001 competition.

A design and analysis of objective function-based unsupervised neural networks for fuzzy clustering

Fuzzy clustering has played an important role in solving many problems. In this paper, we design an unsupervised neural network model based on a fuzzy objective function, called OFUNN. The learning rule for the OFUNN model is a result of the formal derivation by the gradient descent method of a fuzzy objective function. The performance of the cluster analysis algorithm is often evaluated by counting the number of crisp clustering errors. However, the number of clustering errors alone is not a reliable and consistent measure for the performance of clustering, especially in the case of input data with fuzzy boundaries. We introduce two measures to evaluate the performance of the fuzzy clustering algorithm. The clustering results on three data sets, Iris data and two artificial data sets, are analyzed using the proposed measures. They show that OFUNN is very competitive in terms of speed and accuracy compared to the fuzzy c-means algorithm.

A competitive co-evolving support vector clustering

The goal of clustering is to cluster the objects into groups that are internally homogeneous and heterogeneous from group to group. Clustering is an important tool for diversely intelligent systems. So, many works have been researched in the machine learning algorithms. But, some problems are still shown in the clustering. One of them is to determine the optimal number of clusters. In K-means algorithm, the number of cluster K is determined by the art of researchers. Another problem is an over fitting of learning models. The majority of learning algorithms for clustering are not free from the problem. Therefore, we propose a competitive co-evolving support vector clustering. Using competitive co-evolutionary computing, we overcome the over fitting problem of support vector clustering which is a good learning model for clustering. The number of clusters is efficiently determined by our competitive co-evolving support vector clustering. To verify the improved performances of our research, we compare competitive co-evolving support vector clustering with established clustering methods using the data sets form UCI machine learning repository.

Planning based on Dynamic Bayesian Network algorithm using dynamic programming and variable elimination

According to the development of robot technology, Human-Robot Interaction (HRI) is the field of study highlighted. The study aims to find the goal of human action considering their intention and behavior based on their respective habits. To gain the principle of behavior on the goal by understanding that of human, engineers draw the inference of the result needed from Planning through HRI. In this paper, plan inference for aimed goal is modeled by calculating with probability what task system performs through the observed behavior. Dynamic Bayesian Network (DBN) uses the probabilistic inference to reveal the relation of data varying according to time. Machine Repository Pioneer data of UCI has proved that accuracy and efficiency of inference is higher than the existing DBN by lowering useless calculation applying the variable elimination method and the concept of dynamic programming for DBN algorithm.

Discretization of Continuous-Valued Attributes considering Data Distribution

This paper proposes a new approach that converts continuous-valued attributes to categorical-valued ones considering the distribution of target attributes(classes). In this approach, It can be possible to get optimal interval boundaries by considering the distribution of data itself without any requirements of parameters. For each attributes, the distribution of target attributes is projected to one-dimensional space. And this space is clustered according to the criteria like as the density value of each target attributes and the amount of overlapped areas among each density values of target attributes. Clusters which are made in this ways are based on the probabilities that can predict a target attribute of instances. Therefore it has an interval boundaries that minimize a loss of information of original data. An improved performance of proposed discretization method can be validated using C4.5 algorithm and UCI Machine Learning Data Repository data sets.

Unsupervised learning network based on gradient descent procedure of fuzzy objective function

Fuzzy clustering has been playing an important role in solving many problems. Fuzzy c-means (FCM) algorithm is most frequently used for fuzzy clustering. But some fixed points of FCM algorithm, known as Tuckers counter example, is not a reasonable solution. Moreover, the FCM algorithm is impossible to perform online learning since it is basically a batch learning scheme. This paper presents an unsupervised learning network as an attempt to improve the shortcomings of conventional clustering algorithms. This model integrates the optimization function of FCM algorithm into an unsupervised learning network. The learning rule of the proposed scheme is a result of formal derivation based on the gradient descent procedure of fuzzy objective function. Using the result of formal derivation, two implementations of the proposed scheme, the batch learning version and online learning version, are devised. They are tested on Chious data and Iris data and compared with FCM. Experimental results show that the proposed scheme derived the reasonable solution on Tuckers counter example.