An Incremental Multiplexer Problem and Its Uses in Classifier System Research

This paper describes the incremental multiplexer problem, a new test problem for machine learning research. The incremental multiplexer problem is an extended version of the multiplexer problem, a well-known test problem for data mining systems that do classification. The multiplexer problem has been one of the testbed problems driving the development of classifier systems in the last 15 years, but it has one drawback: the number of different multiplexer problems that are interesting and solvable with current computers is very small. This paper describes a generalization of the multiplexer problem that has many instances that are accessible to today's researchers. The paper contains results showing how the performance of a classifier system with fixed parameter settings changes as the size of the message generated by incremental multiplexers increases, suggesting that the incremental multiplexer provides machine learning problems that "fill in the gaps" between the standard multiplexer problems.