Instance based learning approaches for predicting the height of human skeletons

The task of predicting the stature of human skeletal remains using bone measurements is an important one in bioarchaeology. Classical attempts to solve this problem mostly consist of linear regression formulas on various bone lengths. In order to improve these results, we propose using locally-weighted regression and radial basis function networks in order to fit the available data better, especially when using more features and when dealing with complex data that we cannot fit well with linear models. The experiments we performed on a popular data set show that our instance based learning algorithms lead to much better results than the current state of the art. While the methods we propose are computationally expensive by nature, we consider that cost to be irrelevant for the problem at hand, which, according to the existing literature, only ever deals with a few thousand instances at most. Since our methods are almost as easy to apply as linear regression, while providing significantly better results, we consider them very useful for solving such bioarchaeology problems.