In this paper we introduce an approach for selecting a linear model to estimate, in a predictive way, the completion rate of massive open online courses (MOOCs). Data are derived from LMS analytics and nominal surveys. The sample comprises 722 observations (users) carried out in seven courses on EduOpen, the Italian MOOCs platform. We used 24 independent variables (predictors), categorised into four groups (User Profile, User Engagement, User Behaviour, Course Profile). As response variables we examined both the course completion status and the completion rate of the learning activities. A first analysis concerned the correlation between the predictors within each group and between the different groups, as well as that between all the dependent variables and the two response variables. The linear regression analysis was conducted by means of a stepwise approach for model selection using the asymptotic information criterion (AIC). For each of the response variables we estimated predictive models using the different groups of predictors both separately and in combination. The models were validated using the usual statistical tests. The main results suggest a high degree of dependence of course completion and completion rate on variables measuring the user’s behavioural profile in the course and a weak degree of dependence on the user’s profile, motivation and course pattern. In addition, residual analysis indicates the potential occurrence of interaction effects among variables and non-linear dynamics.
Predictive Model Selection for Completion Rate in Massive Open Online Courses / DE SANTIS, Annamaria; Sannicandro, Katia; Bellini, Claudia; Minerva, Tommaso. - In: JE-LKS. JOURNAL OF E-LEARNING AND KNOWLEDGE SOCIETY. - ISSN 1826-6223. - 15:3(2019), pp. 145-159. [10.20368/1971-8829/1135034]
Predictive Model Selection for Completion Rate in Massive Open Online Courses
De Santis Annamaria
;Sannicandro Katia;Bellini Claudia;Minerva Tommaso
2019
Abstract
In this paper we introduce an approach for selecting a linear model to estimate, in a predictive way, the completion rate of massive open online courses (MOOCs). Data are derived from LMS analytics and nominal surveys. The sample comprises 722 observations (users) carried out in seven courses on EduOpen, the Italian MOOCs platform. We used 24 independent variables (predictors), categorised into four groups (User Profile, User Engagement, User Behaviour, Course Profile). As response variables we examined both the course completion status and the completion rate of the learning activities. A first analysis concerned the correlation between the predictors within each group and between the different groups, as well as that between all the dependent variables and the two response variables. The linear regression analysis was conducted by means of a stepwise approach for model selection using the asymptotic information criterion (AIC). For each of the response variables we estimated predictive models using the different groups of predictors both separately and in combination. The models were validated using the usual statistical tests. The main results suggest a high degree of dependence of course completion and completion rate on variables measuring the user’s behavioural profile in the course and a weak degree of dependence on the user’s profile, motivation and course pattern. In addition, residual analysis indicates the potential occurrence of interaction effects among variables and non-linear dynamics.File | Dimensione | Formato | |
---|---|---|---|
DeSantis-PredictiveModelMOOCs.pdf
Open access
Descrizione: File completo con copertine
Tipologia:
Versione pubblicata dall'editore
Dimensione
549.44 kB
Formato
Adobe PDF
|
549.44 kB | Adobe PDF | Visualizza/Apri |
DeSantis-DATASET-PredictiveModelMOOCs.zip
Open access
Descrizione: Dataset
Tipologia:
Altro
Dimensione
437.37 kB
Formato
Zip File
|
437.37 kB | Zip File | Visualizza/Apri |
Pubblicazioni consigliate
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris