This report describes the YAILS[1] incremental learning system. YAILS is a learning by examples system. Its examples and produced rules are described using an attribute-value language. The main characteristics of the program are the following :
# Incrementality
YAILS learns one example at a time, adapting its present knowledge to the new piece of evidence.
# Deals with numerical attributes
Attributes can be declared as numeric. These attributes can originate rules with intervals of values using operators such as >, >=, <, =<, [..], etc.
# Deals with incomplete information (both incomplete descriptions and unknown values)
Examples given to YAILS can be incomplete in the sense that they don't include any information about the value of some attributes. Another source of incomplete information is the possibility of stating that the value of some attribute is unknown.
# Several learning and classifying parameters user-definable
Through the use of several parameters is possible to adapt YAILS to individual problems. Another advantage of this facility is the possibility of orient in some way the main goal of the program either accuracy or comprehensibility.
# Use of a mechanism of controlled redundancy
This is one of the classification features of YAILS which establishes a tradeoff between comprehensibility and accuracy.
# Use of a flexible matching mechanism
Another classification mechanism that enables YAILS to classify examples using rules whose conditions are not completely satisfied. This feature is integrated together with redundancy and conditon's weights.
# Rules conditions differentiated by weights
A mechanism to differentiate the importance of each condition in a rule.