Identify Error-Sensitive Patterns by Decision Tree

TL;DR: This study develops a simple and effective tagging method to digitize individual nodes of a binary decision tree for node-level analysis, to link and track classification statistics for each node in a transparent way, to facilitate the identification and examination of the potentially “weakest” nodes and error-sensitive value patterns in decision trees.

Abstract: When errors are inevitable during data classification, finding a particular part of the classification model which may be more susceptible to error than others, when compared to finding an Achilles’ heel of the model in a casual way, may help uncover specific error-sensitive value patterns and lead to additional error reduction measures. As an initial phase of the investigation, this study narrows the scope of problem by focusing on decision trees as a pilot model, develops a simple and effective tagging method to digitize individual nodes of a binary decision tree for node-level analysis, to link and track classification statistics for each node in a transparent way, to facilitate the identification and examination of the potentially “weakest” nodes and error-sensitive value patterns in decision trees, to assist cause analysis and enhancement development.