歸納邏輯程式設計應用於證券交易相對關係規則之挖掘; Predicate-based relational mining for stock trading

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/13338

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/13338

Title:	歸納邏輯程式設計應用於證券交易相對關係規則之挖掘;Predicate-based relational mining for stock trading
Authors:	莊清焱;Ching-Yen Chuang
Contributors:	資訊管理研究所
Keywords:	相對關係;歸納邏輯程式設計;機率;FOIL;資料挖掘;data mining;Inductive Logic Programming;FOIL;relative comparison;probability
Date:	2007-06-11
Issue Date:	2009-09-22 15:29:23 (UTC+8)
Publisher:	國立中央大學圖書館
Abstract:	本研究運用Muggleton 於1992年提出歸納邏輯程式設計，嘗試解決關於資料挖掘的文獻常使用數值的「絕對」比較，來處理數值型資料方面的問題，事實上使用數值的「相對」比較可以表達的情況會比使用「絕對」比較來得完整。由此可知，如果只使用「絕對」關係挖掘資料，由於背景知識不足的關係，導致學習情況其實有改善的空間。　　本研究將原來的數值比較轉換成邏輯分析，增加「相對」比較的概念，將數值的絕對比較和相對比較當作預測的背景知識，並搭配intensional概念簡化邏輯的描述，設法解決在數值比較的邏輯判斷，會有相同背景知識中item數目過多的問題。　　另外，本研究改良Quinlan於1990年提出歸納邏輯程式設計的FOIL演算法，由於證券市場屬於非結構性的模型，而使用機率性邏輯推理的方式，增加演算法的彈性，使其適合運用在類似證券市場這種沒有清楚定義資料間存在的相關性之模型上，而提出Inductive Probabilistic Programming的概念。　　本研究以學習近日內股價漲跌幅所產生的交易訊號為例，驗證學習正確率及精確率提升的程度，實驗結果證實當加入「相對」比較關係的概念，其學習正確率及精確率會顯著優於只使用「絕對」比較關係來挖掘資料的情況。　　The present research uses the framework of Inductive Logic Programming which is proposed by Muggleton in 1992, and tries to solve the problems which often use the absolute value comparison to handle the numeric data in the previous researches relate to Data Mining. Actually, the situations which use the relative value comparison to express are more complete than to use the absolute value comparison. Due to absolute comparison causes the insufficient background knowledge, we can improve the learning effect of data mining by other suitable techniques. 　　The present research is to transform the original value comparison into logic analysis and increase the concept of relative comparison. It takes the absolute value comparison and relative value comparison as background knowledge of predicate, and collocates the intensional concept to simply the logic description in order to solve that there are many items which represent the same background knowledge in the logic decision of value comparison. 　　Besides, the present research refines the FOIL algorithm of Inductive Logic Programming which is proposed by Quinlan in 1990. Because the stock market is a non-structural model, it has to use probabilistic logic inference to increase the flexibility of algorithm, and let this algorithm fit to apply in the similar model which doesn't define the existent association between data clearly like stock market, so the present research proposes the concept of Inductive Probabilistic Programming. 　　The present research takes stock market as example to learn the trading signals which are caused by the stock price raising or falling several days ago, and verify how much the learning accuracy and precision are improved. The results of experiment confirm when we add the concept of relative comparison, its learning accuracy and precision are obviously better than the situations which only use the absolute comparison.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Size	Format

社群 sharing

Loading...