Classification of Unseen Examples under Uncertainty

Abstract

Very frequently machine learning from real-life data is affected by uncertainty. There are three main reasons for imperfection in data: incompleteness, imprecision (also called vagueness), and errors. In this paper the main emphasis is on classification of unseen examples using a rule set induced from imperfect data. The classification strategy of the machine learning system LERS is described in detail. Results of experiments with medical data sets are also reported.