Recon system, which has been developed to construct the long/short
term portfolios, based on the rule-induction system developed at Lockheed-Martin.In Refs. 3–5, a minimal rule generation and contextual features analysis algorithmfrom IBM research project R-MINI have been presented. It can be observed fromthese research works that combining some data mining techniques with appropriatefinance fundamental factors can produce a promising stock prediction model, whichguarantees high investment return.Classification in the context of data mining is defined as learning a function thatclassifies a data item into one of predefined classes. Classification rules can be consideredas particular kinds of prediction rules. Many classification algorithms and
models have been developed; among them, decision tree is a classical and popularclassification model which was adopted and used in many different areas. From adecision tree model, a set of rules can be produced to make predictions. The C4.5decision tree algorithm,6 a modified version of the ID3 tree algorithm,7 is proven tobe accurate, efficient, and robust by many researches.6,8–10 It is capable of generatingsimple and concise rules while its construction cost is lower than the construction
cost of other computational models such as neural networks and Bayesian inference.In addition, some significant features of the C4.5 algorithm make it suitable for someparticular domains, such as securities analysis, in which data contain a high amount
of noise. For example, it can classify records that have unknown attribute values by
estimating the probability of the various possible results. Moreover, it can deal not
only with the attributes that contain discrete values but also with the attributesthat represent numbers,9 which is very important for security analysis applications.In this work, the C4.5 decision tree model was implemented based on the fundamental
stock data from which a set of stock selection rules was derived forportfolioconstruction. The experimental results, discussed in later sections, show that thegenerated rules have an exceptional predictive performance.The rest of the paper is organized as follows. After discussing the pertinentworks on stock prediction in Sec. 2, the decision tree model construction and ruleA Decision Tree-Based Classification Approach 229extraction procedure are presented in Sec. 3. The rule validation and performanceanalysis are reported in Sec. 4. To further illustrate the applicability of the C4.5decision tree model on the fundamental stock data domain, a second set of rules
was extracted and validated in Sec. 5. Finally, the discussion is given in Sec. 6.
2. Pertinent Works on Stock PredictionIn the past several decades, numerous researches have been done on the stock marketpredication and many techniques and methodologies have been successfully appliedin this area. The statistical-based approaches, neural network-based approaches,
and classification rule-based approaches are the most striking and promising onesamong others.
Statistics has been applied for analyzing the behavior of the stock market formore than half a century. The majorities of the classical statistical stock marketmodels focus on the stock time series prediction and have achieved some acceptableresults. However, due to the random-walk process that the stock market follows andthe non-linearity in the stock data set, the time series models usually cannot reachve
本论文由英语论文网提供整理,提供论文代写,英语论文代写,代写论文,代写英语论文,代写留学生论文,代写英文论文,留学生论文代写相关核心关键词搜索。