Performance analysis of complex valued neural networks for data classification problems

Artificial neural network is frequently preferred method in data processing and classification studies. This method, which is inspired by the transmission model of nerve cells, is a matrix algebra representation that different sequence and connections of neurons generate output. Artificial neural network, which trained according to target output in training step, is applicable for wide range pattern of dataset with variant network topologies. Artificial neural network processing in complex valued geometry is called Complex Valued Neural Network In this study, medical cardiotocography dataset and image skin segmentation dataset for image processing applications, that are accessible in UCI data repository, are tried to be classified by complex valued neural network. Results of prediction accuracy, specificity and sensitivity are used as indicators of model performance over dataset. Outcomes of this thesis show that complex valued neural network model can be applied successfully for cardiotocogram dataset and skin segmentation dataset