Feature Analysis Of Hybrid MOOC Learning Behavior Based On Machine Learning

Massive Open Online Course(MOOC)is booming,it is introduced into classrooms by more and more universities.A hybrid MOOC pedagogy combines traditional closed teaching with open online teaching.In this pedagogy,teaching has been changed,along with the changes of students' learning behaviors.It is different from traditional classroom learning and pure MOOC learning.However,there is little research on learning behavior,especially employing computer technology to analyze the learning behavior in the context of hybrid MOOC.Therefore,this thesis takes MOOC data set as the research object to explore the complex learning behaviors.The MOOC "New Foundation of College Computer" is launched by North China University of Technology,and the data sets are collected from classrooms in the fall semester of 2018.Statistical approaches and machine learning algorithms are employed to extract features of complex learning behaviors.Hybrid MOOC provides a plenty of learning resources with different types.Compared with lecture-style classrooms,Self-Regulated Learning(SRL)behavior comprises a large proportion under this mode.The profile of hybrid MOOC learning behaviors is described from the aspects of knowledge acquisition and application through descriptive statistical analysis.A modeling approach based on grouping with learning attributes of students is designed.The model is applied to explore the differences on the SRL features from various types of students when facing different types of learning tasks.Because video viewing is the main behavior of knowledge acquisition in the context of hybrid MOOC learning,dimension dividing and indicator designing is implemented,ANOVA(analysis of variance)with learning outcome as a grouping variable is employed to compare SRL between groups.The different groups show significant difference on SRL when facing high-order or large-volume tasks,but not significant when facing low-order or small-volume ones.The hybrid MOOC learning behaviors are composed of the ones from multiple dimensions and modes.It is difficult to extract features of learning behaviors in hybrid MOOC through statistical methods in traditional patterns.It is necessary to adopt the clustering algorithm in machine learning algorithm to further classify the complex learning behaviors.In this thesis,a multi-stage dimension reduction algorithm and density peak clustering algorithm are designed to classify the complex learning behaviors in the hybrid MOOC for feature extraction.Considering the multi-dimensional and non-spherical attributes of the current data sets,the clustering algorithm combines Fast Search and Find of Density Peaks(CFSFDP)together with Principal Component Analysis(PCA)and t-distributed Stochastic Neighbor Embedding(t-SNE)dimensionality reduction.The comparison results show that only a single class is identified if employing a single clustering algorithm,and it means no classification.But the combination approach has achieved better results as clustering the data into three different classes.The results show the proposed approach is effective for the classification on complex learning behaviors in the context of hybrid MOOC.