Phenotypic identification of farm animal genetic resources using computer learning with scoring function.

摘要

A precise identification of a given animal as belonging to a given breed is essential for livestock census, and developing policies for selection, improvement and conservation of animal genetic resources. It consists of assigning animals to a breed on the basis of certain phenotypic traits and basically forms a classification problem. Existing computer learning algorithms require a learning data set for predicting the class of a new case. The available information on a breed consists of analysed survey data on a number of its phenotypic traits. Usually this information is presented in the form of breed descriptors as has been prepared for several breeds. This paper reports a learning approach in the form of a scoring function that aggregates trait values to provide a score for identifying breed of the given animal on the basis of breed descriptor. Experiments with the scoring function on both simulated and actual data of four Indian cattle breeds revealed high accuracy of identification. Its performance was comparable to the results obtained with PNC2 (Haendel, 2003, Ph.D. Thesis, University of Dortmund), a recent instance-based learning algorithm. The scoring function technique has been extended to make a decision on breed classification of an animal when breed descriptor of a single breed is available, and also in case the new animal does not belong to any of the breeds under comparison. The technique involves generation of one thousand animals' simulated data from the available breed descriptors and locating the score of new animal in the range of scores of the generated animals for decision support on breed of the new animal.