結合離群值偵測與特徵選取改善預測模型性能;Improving Performance of Prediction Model with Outlier Detection and Feature Selection

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/74662

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/74662

题名:	結合離群值偵測與特徵選取改善預測模型性能;Improving Performance of Prediction Model with Outlier Detection and Feature Selection
作者:	林俊慶;LIN, JUN-QING
贡献者:	資訊工程學系
关键词:	離群值偵測;特徵選取;多元線性迴歸;學習成效預測;Outlier Detection;Feature Selection;Multiple Linear Regression;Learning performance prediction
日期:	2017-07-19
上传时间:	2017-10-27 14:35:28 (UTC+8)
出版者:	國立中央大學
摘要:	為了提升學生的學習成效，提早並準確識別高風險學生，使得教師能夠早期介入輔導，是許多相關研究關注的議題。混成式課程是一種結合線上與線下學習的課程，有別於傳統的線下學習，學生亦能夠透過線上學習平台，來進行多方面的學習。然而，學生在學習過程當中，會留下許多紀錄，例如學生的作業成績、影片瀏覽行為、線上活動頻率、線上測驗成績等等。因此，本論文透過資料探勘與機器學習技術，收集一門混成式微積分課程的學生學習活動資料，使用多元線性迴歸來預測學生的期末成績。相關研究指出，預測模型的準確率容易受到離群值的影響。因此，本論文使用RANSAC演算法，作為離群值偵測的方法，將離群值從資料中去除。為了在移除離群值後更進一步改善預測模型的準確率，本論文以T檢定作為特徵選取的方法，保留對期末成績有顯著影響的關鍵特徵，來進一步改善預測模型的準確率。根據研究結果顯示，透過本論文提出的離群值偵測與特徵選取流程，預測誤差由15.516分降低至4.571分，改善了約70%的預測誤差。 ;In order to improve students’ learning performance, early and accurately identify at-risk students, so that teachers can early intervention, is the focus topic of many related research. Blended course is a course which combine online and offline learning, different from traditional offline learning, students are also able to learn through the online learning platform. However, students will leave a lot of records in the learning process, such as students′ homework grade, video viewing behavior, online activity frequency, online test grade etc. Therefore, this paper based on data mining and machine learning technologies, collects students’ learning activity data from a blended calculus course, uses multiple linear regression to predict students’ final grade. Related researchs point out the accuracy of the prediction model is easily affected by outliers. Therefore, this paper uses RANSAC algorithm as outlier detection method to remove outliers from data. In order to futher improve accuracy of prediction model after remove outliers, this paper uses T-Test as feature selection method, retains the key features that have a significant impact on the final grade, to futher improve accuracy of prediction model. According to the results of research, through the outlier detection and feature selection process proposed in this paper, prediction error from 15.516 down to 4.571 points, improving the prediction error about 70 percent.
显示于类别:	[資訊工程研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	282	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....