Pre-processing Protocol for Nonlinear Regression of Uneven Spaced-Data

Abstract

Regression of experimental or simulated data has important implications in sensitivity studies, uncertainty analysis, and prediction accuracy. The fitness of a model is highly dependent on the number of data points and the locations of the chosen points on the curve. The objective of the research is to find the best scheme for a nonlinear regression model using a fraction of total data points without losing any features or trends in the data. Six different schemes are developed by setting criteria such as equal spacing along axes, equal distance between two consecutive points, constraint in the angle of curvature, etc. A workflow is provided to summarize the entire protocol of data preprocessing, training and testing nonlinear regression models with various schemes using a simulated temperature profile from an enhanced geothermal system. It is shown that only 5% of data points are sufficient to represent the entire curve using a regression model with a proper scheme.