ABSTRACT
This chapter introduces Waikato Environment for Knowledge Analysis (Weka's) MultiLayerPerceptron (MLP) function for supervised learning and Weka's SelfOrganizingMap algorithm for unsupervised clustering. It describes the data sets used for illustrating Weka's neural network function as well as a five-step approach for building feed-forward neural networks. The chapter shows how to use the exclusive-OR (XOR) function to illustrate how MLP builds feed-forward neural networks when the output attribute is numeric. It employs the satellite image data set to revisit attribute selection and to review the process of applying saved learner models to new data of unknown outcome. The chapter describes Weka's implementation of the clustering algorithm to cluster instances of individuals who tested positive or negative for diabetes. The Weka MLP neural network tool allows one to build supervised models for estimation, classification, and prediction. As with MLP, SelfOrganizingMap can be used to cluster data sets having missing values and accepts both categorical and numeric attribute types.
