Learning Functional Descriptors Based on the Bernstein Polynomials - Preliminary Studies

We propose a new method of constructing and estimating descriptors for classifying functional data. These descriptors are based on Bernstein polynomials and their estimation is based on noisy samples of a function (signal) to be classified. The next step is to select an appropriate classifier, well suited to these descriptors. Although the result can be case dependent, we provide the methodology of running comparisons. As a vehicle for presenting the results, we choose benchmark data published in [32]. They represent shocks and vibrations of the operator's cabin in a large mechanical structure.