A Comparative Investigation of Computer Methods for Smoothing and Curve Fitting

Abstract

Seven subroutines for smoothing and curve fitting in FORTRAN and PL/1 were tested and compared for accuracy, speed, and robustness. The smoothing methods were local linear and cubic polynomial fits, and triple exponential smoothing. The curve-fitting methods were approximation by cubic splines, orthogonal polynomials, and rational functions. The data for smoothing and fitting were values of analytic functions with known rounding and random errors. Accuracy of a subroutine was based on the reduction of the original errors. From the comparisons of the results applications of each subroutine were presented in terms of the most suitable types of curves for accurate smoothing and curve fitting.