Akademik Veri Yönetim Sistemi
PAKISTAN JOURNAL OF ZOOLOGY, cilt.50, sa.6, ss.2199-2207, 2018 (SCI-Expanded)
The aim of this study was to develop a honey production model for 180 beekeeping enterprises at Agri, Kars and Erzurum provinces of Turkey and to identify factors affecting honey production through CART, CHAID and MARS data mining algorithms, which are more flexible compared to classical approaches. Several potential predictors in the survey were age of enterprise, province of enterprise (Agri, Kars and Erzurum), educational level, membership status of enterprise to an association of beekeepers (member and nonmember), other activities except for beekeeping (yes and no), number of full beehives, bee race (Caucasian, Carniolan, Italian and Crossbred), and frequency of changing queen bee. MARS outperformed multiple linear regression, and CART in honey yield per hive. No solution for CHAID was generated. In CART algorithm, the highest honey yield per hive (51.250 kg) was obtained from the 33.5 or younger enterprises that performed only beekeeping activity. The four most influential predictors in the MARS were age of enterprise (100%), number of full beehives (100%), other races (97%), and other works except for beekeeping (90%). The best performance order was MARS (r=0.920) > CART (r=0.619) > multiple linear regression (r=0.286), which indicated that MARS outperformed other approaches. MARS reflected that the main and interaction effects of socioeconomic (age of enterprise, province of enterprise, educational level, membership status, other works except for beekeeping and number of full beehives), biological predictors (bee race and frequency of changing queen bee) affected honey yield per hive. As a result, it is recommended that the effect of socioeconomic and biological predictors on the yield should be assessed jointly for further studies.