Abstract
This paper proposes a new type of classifier systems, operon classifier systems (OCS), that develop stable chained classifiers efficiently. To prepare diversity of classifiers that is an essential condition for generating effective chains, OCS is based on niche forming mechanisms such as sharing, covering, and restricted mating. Well-tested chained classifiers are combineed into an operon that tightly maintains subassembries and promotes computational efficiency. Operons keep growing by combining with next families as long as the system performance keeps well. When the performance is aggravated by unsuitable connections, the cut-tail operator that cuts the last-added family from the chain is invoked to regain good performance. OCS is applied to boolean function problems, and results show that OCS develops chained classifiers stably and efficiently.
