Resumen de Understanding and predicting poor performance of computer science students from short time series test results
Chao Mou, Qing Zhou, Xiaoling Zou
Predicting student performance (PSP) is an important application not only for students but also for instructors and staff ineducational institutions. Many educational institutions wouldtest and record studentperformance atregular interval (e.g.,every semester or every year), thus producing a short time series, i.e., a sequence of several data points arranged in temporalorder. Since short time series may not be effectively manipulated with data mining (DM) algorithm, a technique adopted inmost PSP applications, this paper proposes the use of Discrete-time Markov Chain (DMC) and Hidden Markov Model(HMM) for the analysis and prediction of student performance in short time series. This study investigated test results of167 university students with the intention of earning a degree in computer science in the first five semesters since theirenrollment. Analyses showed that HMM can reveal more details than DMC about how student performance changes overtime. Moreover, HMM performs better than DMC and the three commonly used DM techniques in predicting students’performance in the fifth semester from their past test results. HMM was also integrated with clustering technique todetermine the risk levels of poor performance students.
