Machine-Learning Methods on Noisy and Sparse Data

Experimental and computational data field obtained from measurements are often sparse noisy. Consequently, interpolating unknown functions under these restrictions to provide accurate predictions is very challenging. This study compares machine-learning methods cubic splines on the sparsity of training they can handle, especially when samples We compare deviation a true function f using mean square error, signal-to-noise ratio Pearson R2 coefficient. show that, given data, constitute more precise interpolation method than deep neural networks multivariate adaptive regression splines. In contrast, models robust noise outperform after threshold met. Our aims general framework for one-dimensional signals, result complex scientific simulations or laboratory experiments.