Reasoning Methods on Data with Incomplete Description (in Polish)

Rafal Latkowski
Warsaw Univerity, Institute of Computer Science, 2001

Master thesis under supervision of Professor Andrzej Skowron

89 pages, 65 references, 4 tables, 7 figures

Abstract

This work contains an overview of methods that allow reasoning on data with
incomplete object description. Presented here methods consider both, imputing
the missing values, as well as directly reasoning on data with incomplete
object description. The intention of the author was to provide as complete as
possible roadmap of solutions applied in intelligent data analysis and
knowledge discovery together with pointing out their origins. First chapter
introduces the problem of intelligent data analysis using data with incomplete
object description. Second chapter provides basics of Rough-Set theory. Third
chapter presents extension of the Rough-Set theory to the case of missing
attribute values. Fourth chapter presents the inductive learning methods that
deal with problem of missing attribute values beyond the Rough-Set framework.
Fifth chapter describes lazy learning methods dealing with problem of
incomplete object description. Sixth chapter presents imputation methods.
Seventh chapter presents new approach, the Decomposition Method that provides a
possibility to adapt any learning method to the case of missing attribute
values. Presented empirical evaluation shows good efficiency of this method.
Final chapter contains conclusions.