Discovering Interesting Patterns for Investment Decision Making with GLOWER x-A Genetic Learner Overlaid with Entropy Reduction

摘要: Prediction in financial domains is notoriously difficult for a number of reasons. First, theories tend to be weak or non-existent, which makes problem formulation open ended by forcing us consider large independent variables and thereby increasing the dimensionality search space. Second, relationships among nonlinear, may hold only limited areas Third, practice, where analysts conduct extensive manual analysis historically well performing indicators, key find hidden interactions that perform combination. Unfortunately, these are exactly patterns greedy biases incorporated many standard rule learning algorithms will miss. In this paper, we describe evaluate several variations new genetic algorithm (GLOWER) on variety data sets. The design GLOWER has been motivated prediction problems, but incorporates successful ideas from tree induction learning. We examine performance variants two UCI sets as (S&P500 stock returns), using results identify one better further comparisons. introduce (to KDD) (predicting positive negative earnings surprises), experiment with GLOWER, contrasting it tree- rule-induction approaches. Our encouraging, showing ability uncover effective problems have structure significant nonlinearities.