ChaosHunter Compared to Other Modeling Techniques

Top  Previous  Next

Like regression and neural networks, ChaosHunter™ builds equations which best match data and observations. Unlike those more traditional systems, ChaosHunter™  is not bound by any a priori assumptions about the structure of modeling equations. Like regression, it builds “readable” equations, unlike the undecipherable equations neural networks can produce. As a consequence of producing equations more meaningful to humans, these equations sometimes do not match the existing data as closely as neural networks, although ChaosHunter™ often finds underlying relationships that extrapolate better into the “out-of-sample” unseen data. In other words, they are often more robust.


Like genetic programming, ChaosHunter™ uses evolutionary optimization strategies to find modeling equations, which result in some very dynamical forms. Unlike genetic programming, "chromosomes" are a fixed length, so it does not allow chromosomes to be equations or hierarchies of equations. Our method is based on Richard Forsyth's older 1985 methods rather than John Koza's 1988 innovations. However, we believe our method is effective yet inherently reduces the equation complexity, and thus can create robust models with less propensity to "over-fit". In any case, genetic programming is patented, which might subject commercial users to royalty payments.