Ingo Renners

Optimizing fuzzy classifiers by evolutionary algorithms

In this paper a methodology for optimizing fuzzy classifiers based on B-splines by evolutionary algorithms is presented. The algorithm proposed maximizes the performance and minimizes the size of the classifier. On a well-known classification problem the algorithm using only part of the features has a recognition rate comparable to an LDA on the total feature space.

Gaussians-Based Hybrid System for Prediction and Classification

We propose a hybrid model based on Genetic Algorithms (GA), Lattice Based Associative Memory Networks (LB-AMN) and Radial Basis Function Networks (RBFN) for the solution of prediction and classification problems. LB-AMN and RBFN have as basis in their structure a type of asymmetric radial basis function (RBF) which results from the combination of two Gaussian functions. In the first sections we describe the mathematical models used to build the hybrid system. Afterwards, we apply the model to the problem of breast cancer and toxicity prediction. In both cases, the obtained results were better than the ones obtained using other approaches. Finally, some conclusions are given.

Optimizing lattice-based associative memory networks by evolutionary algorithms

Abstract Problem specific network structure optimization subsumes the problem of input selection and network topology identification. Requirements to the network should be accuracy and good generalization abilities. In this contribution we describe in detail an evolutionary algorithm which performs both tasks well. Furthermore, approximation results on mathematical and real world data are presented. In this case we used lattice-based associative memory networks (LB-AMNs) using B-splines as basis functions. The method here is not restricted to B-splines as basis functions. The proposed method and algorithm can be seen as optimized classification system.

Heuristics for Kernels Adaptation in Support Vector Machines

Support Vector Machines are an algorithm introduced by Vapnik and coworkers [9], [10]. They are based on the idea that if input points are mapped to a high dimensional feature space then, a separating hyperplane can be easily found. SVM and kernel methods have been applied to a wide class of problems including approximation and classification and they have proven a remarkable performance on real world problems. An important step in their design is the setting of the kernels parameters which defines the structure of the high dimensional feature space where a maximal margin hyperplane will be found. Too rich feature space, e.g. small kernel parameters, will over-fit the data and hence result in a poor generalisation error, whereas if the kernel parameter is too big, the model will not be able to separate the data. In this paper we firstly propose a heuristic that permits the individual control of the growth in each kernel, which results in more sparse models with higher prediction accuracy. Secondly, a heuristic resulting from the combination of SVM trained by linear programming (LP) and EC for the optimisation of the kernels width is proposed.

Evolutionary System Identification via Descriptive Takagi Sugeno Fuzzy Systems

System identification is used to identify relevant input-output space relations. In this article the relations are used to model a descriptive Takagi-Sugeno fuzzy system. Basic terms of system identification, fuzzy systems and evolutionary computation are briefly reviewed. These concepts are used to present the implementation of an evolutionary algorithm which identifies (sub)optimal descriptive Takagi-Sugeno fuzzy systems according to given data. The proposed evolutionary algorithm is tested on the well known gas furnace data set and results are presented.

Classification Techniques based on Methods of Computational Intelligence

The main focus of this contribution is to present a general methodologyfor the structure optimization of fuzzy classifiers. This approach does not depend on a special type of membership function either it is restricted to small or medium sized input dimension. On a well-known classification problem the algorithm performs an input selection over 9 observed characteristics yielding in a statement which attributes are important with respect to the diagnosis of malignant or benign type of cancer. Results achieved by using different types of basis functions are presented.

Methodology for Optimizing Fuzzy Classifiers Based on Computational Intelligence

In this paper a methodology using evolutionary algorithms is introduced for the optimization of fuzzy classifiers based on B-splines. The proposed algorithm maximizes the performance and minimizes the size of the classifier. On a well-known classification problem the algorithm performs an input selection over 9 observed characteristics yielding in a statement which attributes are important with respect to diagnose malignant or benign type of cancer.

Knowledge discovery with B-spline networks

Due to modern information technology, it is of special interest to consider the analysis of databases in the direction of knowledge discovery. In this paper, we investigate and propose knowledge discovery in the framework of B-spline networks, which are a type of neuro-fuzzy systems. We present an application for knowledge exploration in toxicity prediction by using genetically optimized B-spline networks. A B-spline network can be seen as a classification system which can be applied to knowledge discovery.