Opening the Black Box of Feature Extraction: Incorporating Visualization into High-Dimensional Data Mining Processes

Feature extraction techniques have been used to handle high-dimensional data and experimental studies often show improved classification accuracies. Unfortunately very few studies provide concrete evidences on the effectiveness of these feature extraction techniques and they largely remain to be black boxes. In this study, we design and implement a visualization prototype system that allows users to look into the classification processes, explore the links among the original and extracted features in different classifiers, examine why and how an instance is correctly or incorrectly classified. We demonstrate the prototype's capabilities by combining a feature extraction method based on hierarchical feature space clustering with J48 decision tree classifiers and perform experiments on a real hyperspectral remote sensing image dataset.