Research On Classification Algorithms For Students' Personality Traits

Knowing students' personality traits helps to improve the environment of learning and living in college.There are many new students are asked to complete some kinds of personality questionnaires when they start school.However,the questionnaire which is wildly used now and before,has some inherent limitations such as inefficiency and proneness to cheating.In the last decade,some data mining approaches based on social media sites are proposed to predict people's personality traits.But in college,there is no guarantee that every student frequently uses the same social media site.Thus,we propose a data mining approach based on students' web browsing history and consumption records of campus card to predict student's personality traits.This approach consists of three main parts: feature extraction,feature selection and classifier selection.In the part of feature extraction,we tried to explore students' browsing preference and consumptive habits.Then we removed the redundant features in the part of feature selection.In order to handle the common problem of lack of samples in the area of Educational Data Mining(EDM),a novel approach of adding samples was proposed at the same time.Participants were asked to complete the Revised Eysenck Personality Questionnaire Short Scale for Chinese(EPQ-RSC),and then web browsing history and consumption records were obtained according to students' ID numbers.Finally,the personality traits prediction approach in this paper was applied and we found the best classifier.The experimental results show that the linear SVM performs best with approximate accuracy of 71.8% and F-measure of 77.1% in the dimension of extraversion.What's more,considering of the honesty of participants,this paper made use of the characteristic of EPQ-PSC to select samples.Because of the limitations of Pearson feature filter,this paper used SVM-RFE to improve the feature selection,and found the most important features in each dimension of personality traits.Finally,students' personality traits were both predicted.The linear SVM performed best with 82.5% of accuracy and 86.0% of F-measure in the dimension of extraversion,81.0% of accuracy and 74.4% of F-measure in the dimension of neuroticism,89.0% of accuracy and 88.4% of F-measure in the dimension of psychoticism.The experimental results show that it's possible to use web browsing history and consumption records to predict students' personality traits.This effective objective automatic approach,can detect students' personality traits at any time,which can be wildly used to take place of questionnaires in college.