A Hybrid Evolutionary System for Automated Artificial Neural Networks Generation and Simplification in Biomedical Applications

Data mining and data classification over biomedical data are two of the most important research fields in computer science. Among the great diversity of technique that computer science can use for this purpose, Artificial Neural Networks (ANNs) are one of the most suited. One of the main problems in the development of this technique, ANNs, is the slow performance of the full process. Traditionally, in this development process, human experts are needed to experiment with different architectural procedures until they find the one that presents the correct results for solving a specific problem. However, recently, many different studies have emerged in which different ANN developmental techniques, more or less automated, are described, all of them having several pros and cons. In this paper, the authors have focused to develop a new technique to perform this process over biomedical data. The new technique is described in which two Evolutionary Computation (EC) techniques are mixed in order to automatically develop ANNs. These techniques are Genetic Algorithms (GAs) and Genetic Programming (GP). The work goes further, and the system described here allows the obtaining of simplified networks with a low number of neurons for resolving the problems adequately. Those already existing systems that use EC for ANN development are compared with the system proposed here. For this purpose, some of the most frequently biomedical databases have been used in order to measure the behaviour of the system and also to compare the results obtained with other ANN generation and training methods with EC tools. The authors have also used other databases that are frequently used to compare this kind of method in order to obtain a more general view of the new system’s performance. The conclusions reached from these comparisons indicate that this new system produces very good results, which in the worst case are at least comparable to existing techniques and in many cases are substantially better. Furthermore, the system has other features like variable selection. This last feature is able to discover new knowledge about the problems being solved.