Ordered Estimation of Missing Values for Propositional Learning
Óscar Ortega Lobo, Masayuki Numao · 2000
When trying to discover knowledge by learning concepts embedded in data, it is not uncommon to nd out that the data has missing information. The occurrence of missing information can diminish the condence on the concepts learned from this sort of data. This paper describes a new approach to ll missing values in examples provided to a learning algorithm. In the new approach, a decision tree is constructed to determine the missing values of each attribute by using the information contained in other attributes, ignoring the class. Also, an order for the construction of the decision trees for the attributes is formulated, willing to keep low the computational cost, while still lling important missing attribute values for the classier to be learned later. Experimental results on three datasets show that the approach is successful in providing an input to the decision tree learning algorithm, which leads to nal concepts with less error under dierent rates of rando...