Abstract:
|
In the modern research world, data with functional predictors are increasingly common. We propose a new variable selection technique, when the predictors are functional and the response is scalar. The method is based on a very flexible nonparametric model and we use functional PCA, Gaussian process and similarity regression with L1 penalty to select variables. The method is characterized by high interpretability and computational efficiency. The method is used on hand movement dataset to help enhance health and mobility of lower limb amputees.
|