Research On Student Achievement Prediction Method Based On Mining Of Process-characterised Educational Data

Educational data mining is currently a hot research topic.Predicting student achievement based on accumulated data during student’s learning process is conducive to urging students on daily learning.Also,it is helpful for teachers to discover teaching defects and to make adjustments timely.Traditional methods mostly use educational background,class attendance,class performance,and homework completion quality as the main basis factors for performance prediction.In fact,the process-featured data,such as students’ learning procedures and question answering modes accumulated by online learning platforms,also implies students’ knowledge level information.Therefore,the paper starts from students’ exercise-answering logs accumulated in an online operation platform for course“Program Design Basics",and focuses on the process mining methods of students-shared exercise-answering processes or exercise-answering patterns.Based on the mined results,the paper introduced how to predict student achievement.Detailed work includes the following two aspects:(1)Student-shared exercise-answering process modeling,mining and student achievement grade prediction:Based on students’ submission order of different questions,the judgment results and related time information,this paper proposes the formal model of student-shared exercise-answering processes under specific achievement grades,and develops the corresponding process mining algorithm.Then,the similarity measures between individual student behavior and student-shared exercising processes are presented.After that,the predicting method of students’ achievements grades based on the similarity measures are explored.Two prediction methods based on membership degree vector threshold and on classification learning techniques are presented respectively.(2)Student achievement ranking integrating the feature of students’exercise-answering patterns:This method makes an analogy of student achievement prediction as a search engine ranking problem.First,exercise-answering patterns of a single problem are mined from the answering log.Then,taking both the mined exercise-answering patterns and students’ daily performance before examination as features,the LambdaMart ranking learning algorithm is used to predict the student’s performance ranking in the final examination.The feasibility and effectiveness of the proposed methods are verified with experiments on practical exercise-answering event logs.The experiment result proves the importance of process-featured data for student achievement prediction.It provides a new perspective for student achievement prediction based on educational data mining.