Design of flexible neural trees using multi expression programming

Automatic designing of both architecture and parameters of an artificial neural network is an important problem. This paper introduces a new approach for designing artificial neural networks using multi expression programming (MEP-NN). The approach employs the multi expression programming to evolve the architecture and the parameters encoded in the neural network simultaneously. Based on the predefined instruction sets, a MEP-NN model can be created and evolved. This framework allows input variables selection, over-layer connections and different activation functions for the various nodes involved. The performance and effectiveness of the proposed method are evaluated using stock market forecasting problems and compared with the related methods.