Knowledge Discovery in a Dairy CattleDatabase ( Mining for predictive models )

Proper design of a breeding program has been an issue of primary concern in much animal breeding research during the last decade. Data Mining (DM) is a powerful paradigm for nding patterns that can be used to predict the productivity of progeny given information about their sire, dam and the environment. The more accurate the discovered patterns, the more genetic gain one can achieve in a breeding program. This paper describes a DM process on an Australian dairy database. The focal point of this paper is the selection of a point estimation model for predicting the daughter milk yield within an intelligent decision support system, currently being developed for the Australian dairy industry. The selection of the minimum number of attributes, suucient for satisfactory prediction, and of an accurate mining algorithm form the overall objective of the paper. In addition, the advantages of Bayesian neural networks over conventional feed-forward neural networks are explored.