COMPARISON OF REGRESSION MODELS TO PREDICT POTENTIAL YIELD OF WHEAT FROM SOME MEASURED SOIL PROPERTIES

摘要

Knowledge of potential yield of wheat is imperative for site specific fertilizer management. Data collected from field trials conducted on wheat in Khyber Pakhtunkhwa (KPK) Province, Pakistan was used to predict potential yield of wheat. Various regression models were employed to get the predictions. A complete diagnostic analysis of the residuals of each model is presented. Multiple regression models give us limited prediction power for our data. Models like Classification and Regression Trees (CART) and Random Forests are also explored. The models created are compared on the basis of predictive power and miss-classification error rates. Our results revealed that Random Forests give us very good results if yield is divided into three categories.