結合決策樹與粗略集合論於分析學習行為、學習動機與閱讀素養之關係-以PIRLS2006為例

Analyzing Learning Behaviors, Learning Motivation and Reading Achievement by Decision Tree with Rough Set Theory – A Case Study of PIRLS2006

莊仁翔(Ren- Hsiang Chuang)

指導教授 ： 白炳豐

國立暨南國際大學/管理學院/資訊管理學系/碩士(2011年)

摘要

中文摘要
於2006年台灣參加國際教育成就協會IEA (international association for the evaluation of educational achievement)主導針對國小四年級學生的國際閱讀素養研究PIRLS (progress international reading literacy study)，報告結果顯示台灣與其他44國家在國際閱讀素養研究中發現台灣成績在全部45個國家排名第22名，因此，本研究希望可以透過資料探勘分析技術來探討影響台灣學生閱讀素養的主要因素以及提供促進學生閱讀素養的建議途徑。我們在研究中發現高學習動機與高頻率的學習行為對閱讀成就具有正向的影響，學習動機因子如同問卷選項的“當我想要時才閱讀”、 學習行為因子如同問卷選項的“我靜靜的自己閱讀”以及閱讀成就依國際標準分成頂標(625)、高標(550)、中標(475)與低標(400)。
本研究提出混合C5.0RST分析模組(C5.0決策樹和RST粗略集合論)運用於分析台灣國小四年級學生的閱讀素養，提供使用者了解學習影響因素以及使用者也可以透過輸入相關因素，進而獲得改進建議途徑，假如需改進因素多於2個，則提供影響因素權重排序，C5.0決策樹主要進行資料前處理部分，藉此產生重要屬性挑選與屬性權重排序的結果，而RST粗略集合論方法則是利用已前處理過的資料建置規則庫，提供決策者清楚了解造成事件發生的原因進而下決策。
實驗結果顯示本研究提出的C5.0RST分析模組可以達到良好的分析結果，並且在與其他屬性挑選方法比較後，如判別分析(LDA)、ID3決策樹(ID3)與C4.5決策樹(C4.5)，皆顯示本研究所提出的C5.0RST分析模組提供良好的分析結果與縮短分析時間。
關鍵字：閱讀素養、資料探勘、C5.0決策樹(C5.0)、粗略集合論(RST)、判別分析(LDA)、ID3決策樹(ID3)、C4.5決策樹(C4.5)。

關鍵字

閱讀素養 ； 資料探勘 ； C5.0決策樹(C5.0) ； 粗略集合論(RST) ； 判別分析(LDA) ； ID3決策樹(ID3) ； C4.5決策樹(C4.5)

並列摘要

Abstract
The study proposed a C5.0RST model incorporating by C5.0 decision tree and rough set theory to analyze reading literacy data from the IEA(international association for the evaluation of educational achievement) PIRLS(progress international reading literacy study) 2006 based on reading literacy scores and student questionnaire data. The C5.0 decision tree was utilized to determine the important influential factors for reading literacy as well as enhance the efficiency of RST. The PIRLS leading by the IEA in 2006, Taiwan was ranked 22 among the 45 countries. The result was merely above the average. Thus, the assessment of reading literacy of primary school pupil’s learning performance becomes the most essential issues of the educational research in Taiwan. In our research, we found high degree of learning motivation and high frequencies of learning behaviors are positive effect on reading achievement. Learning motivation factor as like questionnaires “I read only if I have to”, learning behaviors factor as like questionnaires “I read silently on my own” and reading achievement corresponds to international benchmark (as like advanced international benchmark (over 625), high international benchmark (over 550), intermediate international benchmark (over 475) and low international benchmark (over 400)). The informative factors were reached from the primary school student questionnaires by C5.0 decision tree, and then the selected features provided for RST to derive comprehensive decision rules. The result could give a suitable direction for educationist to make a proper decision. Moreover, to robust the feature selection procedure, we applied three other techniques namely, LDA, ID3, C4.5, to compare the feasibility. Finally, the ANOVA test was applied to assess the significant of the decision rules.
Key Word: Reading Literacy, C5.0 decision tree (C5.0) , Rough set theory (RST), Linear discriminant analysis (LDA), ID3 decision tree (ID3), C4.5 decision tree (C4.5).

並列關鍵字

Reading Literacy ； C5.0 decision tree (C5.0) ； Rough set theory (RST) ； Linear discriminant analysis (LDA) ； ID3 decision tree (ID3) ； C4.5 decision tree (C4.5)

參考文獻

Reference

[1] Alan Talevi, Mariana Sella Cravero, Eduardo A. Castro, and Luis E. Bruno-Blanch, (2007). Discovery of anticonvulsant activity of abietic acid through application of linear discriminant analysis, Bioorganic & Medicinal Chemistry Letters 17, 1684–1690.

[2] Ann Kennedy, Examining Gender and Fourth Graders’ Reading Habits and Attitudes in PIRLS 2001 and 2006.< http://www.iea.nl/irc2008_pirls.html>

[3] Chu, S., Tse, S.K., Loh, E. K. Y., Chow, K., Fung, H. F., and Ng, H. W. R. (2008), Primary four students’ development of reading ability through inquiry-based learning projects, The 2008 IASL Conference [CD-ROM], University of California at Berkeley, Berkeley, CA, USA, 3rd to6th August 2008.

[4] D. Pomorski and P.B. Perche (2001). Inductive learning of decision trees: application to fault isolation of an induction motor. Engineering Applications of Artificial Intelligence 14, 155-166.