Both K-nearest neighbor and support vector machine regressors are distance-based predictors.
The support vector regressor with a linear kernel,
linearSVR, somewhat mimics the linear regression results above.
SVR with a polynomial kernel,
polySVR, converges to
linearSVR in performance as degree of the polynomial kernel is reduced to 1. The
sigmoidSVR using a sigmoid kernel is the most inferior of the
SVR model with a radial-basis-function,
rbfSVR, produces the best performance with
R²=0.88 and std
0.34. This is the best result achieved in this study.
The K-neighbors regressor performance is also very strong.