Knowledge guided evolutionary algorithms in financial investing.

详细信息   

• 作者：Wimmer ; Hayden.
• 学历：Ph.D.
• 年：2013
• 导师：Rada, Roy,eadvisorRada, Royecommittee memberYoon, Victoriaecommittee memberZhou, Binecommittee memberJaneja, Vandanaecommittee memberZhou, Linaecommittee member
• 毕业院校：University of Maryland
• Department：Information Systems
• ISBN：9781303393037
• CBH：3594636
• Country：USA
• 语种：English
• FileSize：4120612
• Pages：206

文摘

A large body of literature exists on evolutionary computing, genetic algorithms, decision trees, codified knowledge, and knowledge management systems； however, the intersection of these computing topics has not been widely researched. Moving through the set of all possible solutions&mdash；or traversing the search space&mdash；at random exhibits no control over how an organism evolves into a new organism that lies within the search space, also known as the mutation process. Employing codified human knowledge, the traversal of the search space may be constrained and directed thereby controlling mutation. This reveals the research question: How can codified human knowledge be integrated into the evolutionary mutation process in order to influence the traversal of the search space In order to investigate the research question, a design science approach will begin with a design artifact by creating DEFINE&mdash；Dynamic Evolution in Financial Investment Election. DEFINE employs decision trees as organisms and utilizes ontologies to constrain the mutation process. This research has the potential to reveal new and unexplored methods of employing human knowledge coupled with evolutionary computing to produce quality knowledge management systems. Preliminary experiments reveal an advantage, as measured by increased classification accuracy, to incorporating knowledge in the mutation process and also reveal financial markets follow a path of evolution with less than 10 generations to reach optimal fitness. Additional experiments conclude well-constructed knowledge constrains the mutation process to exercise control over the traversal of the search space whereas poorly-formed or random knowledge has no effect. Employing knowledge into the mutation process demonstrates making small changes in the organisms structure leads to small changes in the organisms function； however, making large changes showed mixed results. Performance, as measured by classification accuracy and number of generations to reach an optimum fitness, is improved by making small changes in the organism structure as opposed to random changes. This research demonstrates the number of years in time series data affects the number of generations required to achieve an optimum fitness. The methods explored in this research are applied to a financial investing scenario and profitability is shown to lie between two common investment strategies.