Picturing Bayesian Classifiers: A Visual Data Mining Approach to Parameters Optimization

The goal of supervised classification is to assign a new object to a class from a given set of classes based on the attribute values of this object and on a training set. Although "supervised," classification algorithms provide only very limited forms of guidance by the user. Typically, the user selects the dataset and sets the values for some parameters of the algorithm, which are often difficult to determine a priori. We believe the user should be involved more interactively in the process of classification because, by providing adequate data and knowledge visualizations, the pattern recognition capabilities of the human can be used to increase the effectivity of classifier construction. Moreover, users often want to validate and explore the classifier model and its output. To address these issues, the classification system should have an intuitive and interactive explanation capability. We present a two-dimensional visualization tool for Bayesian classifiers that can help the user understand why a classifier makes the predictions it does given the vector of parameters in input. The user can interact with the classifier by: selecting different models and changing the parameters of the prior. To help people discover (sub)optimal parameters, we develop a visual interaction method that allows objects to be interactively analyzed. Finally, we present a case study to demonstrate the effectiveness of our solution in text classification. © 2014 Elsevier Inc. All rights reserved.