DATABASE MINING OF TURFGRASS CULTIVAR PERFORMANCE

Turfgrass quality is,generally determined by researchers based on thei r previous experience. This evaluation mechanism has been used to,gath er performance data on a large collection of turfgrass cultivars in Na tional Turfgrass Evaluation Program (NTEP) studies across the United S tates. NTEP collects data, evaluates it, and distributes the findings to turfgrass managers and educators. We developed an automated procedu re for exploring the NTEP database to discover instances of high culti var quality or performance in unique,growing conditions. Ln previous w ork, a procedure was developed to pre-process the data to normalize di fferences among evaluators and the C4 learning algorithm was selected as the most useful for automatically learning rules that identified th e unique,growing conditions for high quality cultivars. This study exa mined only a small portion of the NTEP database. C4 discovered 85 rule s by examining data collected for 20 Kentucky bluegrass (Poa pratensis L.) cultivars over an eight-year period. Of these 85 rules, 71% fit t he accepted management and environmental parameters for Kentucky blue- grass. The remaining 29% of the rules showed unexpected relationships among management and environmental parameters. A best-management exper t system for turf would require a more comprehensive examination of th e NTEP database along with many additional rules covering turf managem ent operations.