Results

Due to its ability to identify and focus on driving variables, Symbolic Regression can build models from data sets that have more variables than records (these are commonly known as fat arrays, and most non-evolutionary machine learning techniques find issues to deal with them) \cite{2016}. Therefore, we apply genetic programming for dimensionality reduction purposes, and to build the predictive models that can provide insight into the shape of the trust data space.

Bitcoin

In total, 532 model were generated, with the majority of those (55.1%) containing at least three variables.  The modeling process explores the trade-off between ModelComplexity and model error (1-R^2). This is illustrated in the ParetoFrontLogPlot which displays each of the returned models' quality metrics, complexity and accuracy. The models denoted by red dots are all optimal in the sense that for a given level of accuracy there is no simpler model or, conversely, for a given level of complexity there is no more accurate model \cite{Kotanchek}. Notably, there are 3 models at an order of magnitude  materially better than the rest (error on a scale under 10^-15), and one of those is an optimal model .