Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance

Rianne Conijn, Ad Kleingeld, Uwe Matzat, Chris Snijders, Menno van Zaanen

    Research output: Contribution to conferencePaperOther research output

    142 Downloads (Pure)

    Abstract

    The use of learning management systems (LMS) in education make it possible to track students’ online behavior. This data can be used for educational data mining and learning analytics, for example, by predicting student performance. Although LMS data might contain useful predictors, course characteristics and student characteristics have shown to influence student performance as well. However, these different sets of features are rarely combined or compared. Therefore, in the current study we classify student performance using information from course characteristics, student characteristics, past performance, and LMS data. Three classifiers (decision tree, rule-based, and SVM) are trained and compared with the majority class baseline. Overall, SVM is the best classifier to identify pass/fail for a student in a course. However, for more interpretable results, the decision tree or the rule-based algorithm with course characteristics, student characteristics, and midterm data are good second bests. Additionally, it is shown that the different feature sets all have a positive influence on predicting pass/fail. In particular, student characteristics and the midterm grade have a large influence. Compared to these feature sets, LMS data seems less important. Yet, a more fine-grained
    analysis of the specific metrics found in the learning management system may still yield useful information.
    Original languageEnglish
    Publication statusPublished - Dec 2016
    EventNeural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016 - Barcelona, Spain
    Duration: 10 Dec 201610 Dec 2016
    http://ml4ed.cc/2016-nips-workshop/

    Conference

    ConferenceNeural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016
    Abbreviated titleNIPS 2016 ml4ed
    CountrySpain
    CityBarcelona
    Period10/12/1610/12/16
    Internet address

    Fingerprint

    Students
    Decision trees
    Classifiers
    Data mining
    Education

    Cite this

    Conijn, R., Kleingeld, A., Matzat, U., Snijders, C., & van Zaanen, M. (2016). Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance. Paper presented at Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016, Barcelona, Spain.
    Conijn, Rianne ; Kleingeld, Ad ; Matzat, Uwe ; Snijders, Chris ; van Zaanen, Menno. / Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance. Paper presented at Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016, Barcelona, Spain.
    @conference{35cd6ad9004b410588d7fefa602b9e99,
    title = "Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance",
    abstract = "The use of learning management systems (LMS) in education make it possible to track students’ online behavior. This data can be used for educational data mining and learning analytics, for example, by predicting student performance. Although LMS data might contain useful predictors, course characteristics and student characteristics have shown to influence student performance as well. However, these different sets of features are rarely combined or compared. Therefore, in the current study we classify student performance using information from course characteristics, student characteristics, past performance, and LMS data. Three classifiers (decision tree, rule-based, and SVM) are trained and compared with the majority class baseline. Overall, SVM is the best classifier to identify pass/fail for a student in a course. However, for more interpretable results, the decision tree or the rule-based algorithm with course characteristics, student characteristics, and midterm data are good second bests. Additionally, it is shown that the different feature sets all have a positive influence on predicting pass/fail. In particular, student characteristics and the midterm grade have a large influence. Compared to these feature sets, LMS data seems less important. Yet, a more fine-grainedanalysis of the specific metrics found in the learning management system may still yield useful information.",
    author = "Rianne Conijn and Ad Kleingeld and Uwe Matzat and Chris Snijders and {van Zaanen}, Menno",
    year = "2016",
    month = "12",
    language = "English",
    note = "Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016, NIPS 2016 ml4ed ; Conference date: 10-12-2016 Through 10-12-2016",
    url = "http://ml4ed.cc/2016-nips-workshop/",

    }

    Conijn, R, Kleingeld, A, Matzat, U, Snijders, C & van Zaanen, M 2016, 'Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance' Paper presented at Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016, Barcelona, Spain, 10/12/16 - 10/12/16, .

    Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance. / Conijn, Rianne; Kleingeld, Ad; Matzat, Uwe; Snijders, Chris; van Zaanen, Menno.

    2016. Paper presented at Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016, Barcelona, Spain.

    Research output: Contribution to conferencePaperOther research output

    TY - CONF

    T1 - Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance

    AU - Conijn, Rianne

    AU - Kleingeld, Ad

    AU - Matzat, Uwe

    AU - Snijders, Chris

    AU - van Zaanen, Menno

    PY - 2016/12

    Y1 - 2016/12

    N2 - The use of learning management systems (LMS) in education make it possible to track students’ online behavior. This data can be used for educational data mining and learning analytics, for example, by predicting student performance. Although LMS data might contain useful predictors, course characteristics and student characteristics have shown to influence student performance as well. However, these different sets of features are rarely combined or compared. Therefore, in the current study we classify student performance using information from course characteristics, student characteristics, past performance, and LMS data. Three classifiers (decision tree, rule-based, and SVM) are trained and compared with the majority class baseline. Overall, SVM is the best classifier to identify pass/fail for a student in a course. However, for more interpretable results, the decision tree or the rule-based algorithm with course characteristics, student characteristics, and midterm data are good second bests. Additionally, it is shown that the different feature sets all have a positive influence on predicting pass/fail. In particular, student characteristics and the midterm grade have a large influence. Compared to these feature sets, LMS data seems less important. Yet, a more fine-grainedanalysis of the specific metrics found in the learning management system may still yield useful information.

    AB - The use of learning management systems (LMS) in education make it possible to track students’ online behavior. This data can be used for educational data mining and learning analytics, for example, by predicting student performance. Although LMS data might contain useful predictors, course characteristics and student characteristics have shown to influence student performance as well. However, these different sets of features are rarely combined or compared. Therefore, in the current study we classify student performance using information from course characteristics, student characteristics, past performance, and LMS data. Three classifiers (decision tree, rule-based, and SVM) are trained and compared with the majority class baseline. Overall, SVM is the best classifier to identify pass/fail for a student in a course. However, for more interpretable results, the decision tree or the rule-based algorithm with course characteristics, student characteristics, and midterm data are good second bests. Additionally, it is shown that the different feature sets all have a positive influence on predicting pass/fail. In particular, student characteristics and the midterm grade have a large influence. Compared to these feature sets, LMS data seems less important. Yet, a more fine-grainedanalysis of the specific metrics found in the learning management system may still yield useful information.

    M3 - Paper

    ER -

    Conijn R, Kleingeld A, Matzat U, Snijders C, van Zaanen M. Influence of course characteristics, student characteristics, and behavior in learning management systems on student performance. 2016. Paper presented at Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Education 2016, Barcelona, Spain.