Scientific publications

Submitted

Abstract: We will describe how neural networks combined with genetic algorithms can be used for building a bankruptcy prediction model. We start with a short description of the theory behind neural networks and genetic algorithms. Then we describe a typical neural network model building procedure, i.e. we present the data, split it in three parts, choose the input variables, train the networks and test the networks. Finally, we compare the results with results obtained using statistical methods. We use a data set that contains financial data on 570 failed and nonfailed companies included in Compustat in 1985-1993.

In conference proceedings

Abstract: This paper presents an empirical comparison of four evolutionary encoding methods for finding suitable neural network topologies. We use five different learning problems for benchmarking the encoding methods. Three of the problems are artificial (Encoder and two function approximation problems), and two are real-world classification problems from the PROBEN1 benchmarking problem set. We train the network weights with a separate neural learning algorithm. Our evaluation criteria are classification accuracy and efficiency for using only the relevant variables.

Thesis

Abstract: This thesis compares some methods for finding neural network architectures suitable for learning particular problems. We use an evolutionary algorithm with four different genetic encoding methods to search for the suitable architectures. We train the neural network weights with a separate neural learning algorithm. We use eight different learning problems for benchmarking the encoding methods. Four of the problems are artificial (XOR, Encoder and two function approximation problems), three are real-world classification problems from the Proben1 benchmarking problem set, and one is a bankruptcy classification problem studied earlier in one of our projects. Our evaluation criteria are classification accuracy and efficiency for using only the relevant variables. The classification results are compared also to those for network architectures found by a systematic search.

Compressed PostScript (1018 kB, .ps.gz)
Bibliography entry (.bib)
Thesis homepage, including all source code
Papers might not always be available online!