PREDICTIVE MODELLING AND VALIDATION FOR ESTIMATING FODDER YIELD OF GREWIA OPTIVA

Abstract

Grewia optiva is one of the most important fodder trees of north-western and central Himalayas. It provides fodder during lean period of winter when there is scarcity of other fodders. For modelling, regression analysis was used to study the relationship between fodder yield (dependent variable) and other parameters. In total, more than 30 models (including linear and non-linear) were tried and on the basis of adjusted R2, the best five models were selected. These five models were validated for its adequacy through different criteria, namely, adjusted R2, bias, variance, root mean square error and coefficient of dispersion. On the basis of set criteria, the models were ranked. After applying the Wilcoxon signed rank test on fitting data set, one can arrive at the final ranks by considering ranks of both fitting (Rf) and validating (Rv) data sets. Finally, on the basis of all the criteria adopted in the present investigation, out of the best five models, the regression model obtained as  =  8.467 + 0.000004 (L2*S) ranked first, where  = estimated fodder yield, L = average number of leaves per secondary branch (S), and hence recommended for fodder yield prediction of Grewia optiva.   Keywords: Grewia optiva, Modelling, Validation, Rank.