Enhancing the Predictions Accuracy of Bovine Lameness Models through Nonlinear Transformations of the Limb Movement Variables

This paper demonstrates that the prediction accuracy of bovine lameness models improves when the explanatory variables, i.e. limb movement variables (LMV) used in the logistic regression functions are transformed via nonlinear transformations. The nonlinear transformations are performed using 2 main parameters: the number of knots and the degree of freedom. The former impacts the accuracy of the lameness predictions immensely, whereas the latter has no effect on model accuracy.
It is further demonstrated that prediction accuracy and system robustness are diverging as the number of knots surpasses certain level. System robustness is defined as lameness prediction accuracy when the LMV are perturbed due to measurement noise. The optimal number of knots presented here was derived by computer simulations. Experimental validations of this number in farm settings are to follow ...