Data Mining of Subjective Agricultural Data

Abstract

The National Turfgrass Evaluation Program was established to collect data on the performance of cultivars of different turfgrass species. This data is highly subjective, consisting of visual ratings of turfplots by evaluators of varying experience at sites throughout the U.S. In this paper we describe a methodology for mining this large database through learning models to make recommendations about specific cultivars depending upon maintenance and environmental conditions. Our experiments show that learned models outperform human experts in a number of areas. Our studies also indicate that the many of human experts lack the necessary experience to discriminate between specific cultivars. This knowledge is of great commercial and environmental importance in the turfgrass industry.