Strategies for managing missing and incomplete information with applications to keystroke biometric data and a business analytical application
Missing Data Techniques (MDTs) are examined and placed into context by applying them to a keystroke biometric application and database, and then applying our findings to the broader topic of analytical applications. Multiple MDTs are examined in light of the nature and the amount of missing data and applied to the keystroke system through the vehicle of fallback models, a technique inspired by research in speech recognition. We conclude that heuristic-based MDTs are more effective than statistic-based techniques, that multiple imputations are more effective than single imputations, and that effectiveness overall deteriorates as the amount of missing data increases. Through the application of MDTs instantiated through fallback models, we were able to improve the performance of the keystroke system slightly, as determined by chi square analysis. Lastly, we apply our findings to the growing area of analytical applications and suggest a new, more thorough model in which the user is made aware of the nature and amount of missing data in a given data set, and is able to better and more intelligently manage that missing data through the selection of the most appropriate MDT. We develop and illustrate this new model through the creation of a feasibility simulation.^
"Strategies for managing missing and incomplete information with applications to keystroke biometric data and a business analytical application"
(January 1, 2010).
ETD Collection for Pace University.