Created by W.Langdon from gp-bibliography.bib Revision:1.5356
Predictions are available to users from different sources. We investigate whether FGP-1 has the capability of improving on them by combining them together. Based on the experiments presented in this thesis, we conclude that FGP-1 is capable of improving the given predictions in terms of prediction accuracy. This partly attributes the capability of FGP-1 of finding positive interactions between the predictions given. However, caution should be excised for choosing its parameters in such applications.
Improving prediction precision is highly demanded in financial forecasting. Our investigation is based on a set of specific prediction problems: to predict whether a required rate of return can be achieved within a user-specified period. In order to improve prediction precision, without affecting the overall prediction accuracy much, we invent a novel constrained fitness function, which is used by FGP-2. The effectiveness of FGP-2 is demonstrated and analysed in a variety of prediction tasks and data sets. The constrained fitness function provides the user with a handle to improve prediction precision at the price of missing opportunities.
This thesis reports the utility of FGP applications to financial forecasting to a certain extent. As a tool, FGP aims to help users make the best use of information available to them. It may assist the user to make more reliable decisions that would otherwise not be achieved without it.",
Genetic Programming entries for Jin Li