Creating a Comprehensive Feature Space Library for Machine Learning
Infotech@Aerospace 2012 · 2012
The practical issue of ensuring comprehensive feature space is available for machine learning in specific domains has not received sufficient attention. In large multivariate data sets, it is not difficult to collect overwhelming numbers of examples from aerospace systems or other sources. However, just because a large number of examples are present does not ensure that the full feature space experienced by the system is represented. Without a methodology to objectively account for a comprehensive feature space, the training sets for machine learning are likely to be incomplete and may not adequately represent the features that are desired to be learned. This paper addresses the subject and provides insight into a practical methodology to start accounting for the comprehensive universe of feature space examples for specific domains as well as addressing relevance to the objective of data-driven computational