利用模糊類神經網路模型進行學習成效預估

Estimation of Learning Effectiveness by Fuzzy Neural Network Models

林家聖(Chia-Sheng Lin)

指導教授 ： 賀嘉生

中原大學/工學院/資訊工程研究所/碩士(2005年)

https://doi.org/10.6840/cycu200500047

摘要

資料探索的工作在世界上已行有多年，許多的研究人員在這個領域裡努力地想要挖掘出更多更有用的資訊出來，這些資訊往往是一些關聯或是規則。透過這些被挖掘出來的關聯或規則，它們告訴了人們這些資料在過去的歲月中曾經存在著哪些的模式。除此之外，資料裡面的這些模式更隱含了人們所希望知道的未來，接著資料預測的研究就漸漸地展開了起來。只不過這些資料預測的研究大多專注於商業活動之中，例如股票走勢，需求預測等等，當然非商業行為的預測還是存在著，比如說氣象預測或是應用於網路快取上的預測都有，只不過針對教育學習的研究卻是屈指可數。
本論文之中利用了類神經網路裡頭的自組織圖，結合資料探索領域當中的資料聚類(Clustering)理論，設計了三種資料預測模型。包括狀態轉換矩陣、狀態轉換機率模型以及連續狀態轉換模型等等，企圖透過時序型資料之中資料的形態(pattern)轉換來進行資料預測。在這一個研究之中，待預測資料會透過資料預測模型的使用，找出未來可能的落點，這些落點將會被模糊理論的方法給模糊化之後，萃取出這些它們隱含的資訊，最後這些資訊將告訴我們未來資料落在某些系統預設落點上的機率有多高，產生一種較為制式化的預測結果報告。
除了資料的預測方法之外，本論文裡頭提出了一種全迴路的資料處理系統，結合上述的資料預測方法，並應用於學生學習成績的預測之上。全迴路系統資料有如用一般標準資料處理的程序之外，還會對預測的結果進行分析，最後回饋到學生身上，完成一段完成的迴路。在本論文的實驗中，我們分別以一個、兩個到三個資料預測模型的應用，進行學生學習成績的預測，看看是否在更多預測模型的合作之下，能夠找出更準確的結果，除此之外也利用了不同學院的學生成績進行預測，以確保資料預測方法的一般性。

關鍵字

資料預測 ； 模糊理論 ； 資料探索 ； 類神經網路 ； 時序型資料

並列摘要

Research in data mining domain have a long period, many researcher works hard for digging more and useful information, which are some kind of relationship or rules. But with these reslationship and rules, they tell people what happened in and if there are some behavior patterns in the past. Besides, these patterns not only being patterns but also tell people where the future is. These patterns bring the age of research in estimation, but most of these researches pay too much attention on business activities, such as stock trend, demand forcasting etc. Some researchs in non-business activities are exist surely, for example the predition of climate or cache which applied to internet. But only few researches focused on education.
In this thesis, we have designed 3 kinds of data estimating model, with the data clustering theory in data mining domain, by using the self-organizing map (SOM) of neural network. These models are state-transition matrix, state-transition probabilitic model and continuous state-transistion model, which use in serial data for finding internal pattern transition. By using of these these models, it gets some possible value in the future, and these values will be extracted for the inner information by fuzzy theory. With the information, we can find the probabilities of some default values, for a formal report of estimation.
This theis brings up not only a estimation method, but also an full-loop data mining process. The process integrates with the estimation method which has mentioned above, it applies to the the estimation of the l nmg effectiveness. Full-loop data mining process is like most KDD processes, from original data to model, but the process also uses these models for analyzing the grades to students for a feedback. Experiments in the thesis predict the values of fall of students by using 1 to 3 models, to confirm if the estimation more accurate by the cooperation of more models. Beside, for confirming the generality, we also used the data of student grades in different college.

並列關鍵字

fuzzy theory ； data serial ； neural network ； data mining ； estimation

參考文獻

[黃天鴻03] 黃天鴻，「以知識地圖為基礎發展的網路全迴路學習」，中原大學資訊工程研究所碩士論文，2003

[CMS71] John C. Champer, Satinder K. Mullick, and Donald D. Smith, “How to Choose the Right Forecasting Technique,” Harvard Bussiness Review, July-Aug. 1971, pp.45-74.

[Aok79] Takashi Aoki, “A statistical prediction of the tropical cyclone position based on persistence and climatological factor in the western North Pacific (the PC method),” Geophys. Mag., Japan Meteoro-Logical Agency, No.4. 1979

[BaP92] H.-U. Bauer and K. Pawelzik, “Quantifying the neighborhood preservation of self-organizing feature maps”, IEEE Trans. Neural Networks, Vol.3, No.4, 1992, pp-570-579.

[BaW03] Arvind K. Bansal and Cristopher J. Woolverton, “Applying Automatically Derived Gene-Group to Automatically Predict and Refine Metabolic Pathways,” IEEE Transactions on Knowledge and Data Engineering, 2003, pp.883-894.