Applying fuzzy clustering to diagnose and modify neural network models in financial engineering

The paper illustrates the application of fuzzy clustering to interpret the performance of a neural network model making price predictions. The goal is to identify what particular combinations of input variables lead to correct predictions by the model and which do not. This knowledge may be used to alter financial trading actions based on model predictions by conditioning those decisions on the current state of input variables. Alternatively, these results may be used to direct changes to the fundamental architecture of the original neural network model. Fuzzy c-means (FCM) clustering is shown to successfully provide this knowledge.