An application of predicting student performance using kernel k-means and smooth support vector machine

This thesis presents the model of predicting student academic performances inHigher Learning Institution (HLI).The prediction ofstudentssuccessfulis one of the most vital issues inHLI.In the previous work, thereare many methodsproposed topredictthe performanceof students such as Scholastic Aptitude Test (SAT) or American College Test (ACT), Intelligent Test, Fuzzy Set Theory, Neural Network, Decision Tree and Naïve Bayes.However, thefactremainsfound ina variety of debateamongeducators inhigher learning institution, especially those relatedto predictorvariablesthatused and the resulting level of prediction accuracy.This shown that the rule model in predicting student performanceisstilla gapand it is urgent for educators to obtain a more accurate prediction results.The objective of thisstudyis to create a rule model in predicting of students performance based on their psychometric factors. In this study, psychometric factors used as predictor variables, thereare Interest, Study Behavior, Engaged Time, Believe, and Family Support.The rulemodel developed using Kernel K-means Clustering and Smooth Support Vector MachineClassification.Both of these techniquesbased on kernel methodsand relativelynew algorithms of data mining techniques, recently received increasingly popularity in machine learning community. These techniques successfullyapplied in processing large amounts of data, especially on high dimensional data that are nonlinearly separable. The data collection from student academic databases and surveyed the psychometric factors of undergraduatestudentin semester 3 sessions 2007/2008 at Universiti Malaysia Pahang.Theresultof this study indicatesa positive correlation between the proposed predictor variables and the students performance.These predictor variables contributesignificantly in increasing or decreasing student performance that is equalto52.2%(R2=0.522).The studyalsofound the cluster model of students based on their performance. Eachmember of the clusters labeledwith their performance index to describe the current condition of student performance.The prediction accuracy of predicting modelproposed have thelowest accuracy 61%(R2= 0.61)in predicting Good performance indexand thehighest accuracy 93.67% (R2= 0.9367)in predicting Poor Performance index. This studyshowedthat the kernel methodhasa capabilityas data mining technique on educational data mining. The results of this studyaresuitableto beusedinmonitoringthe progression of students performancesemester by semesterand supportedthe decision making process by decision makerinHLI.