Prediction of Outcome in Massive Open Online Course by making use of various Machine Learning classification algorithms.
Massive Open Online Courses (MOOCs) have the potential to enable free university-level education on an enormous scale.Coursera, edX, and Udacity offer MOOCs though some are being offered directly by a college or university. Student interaction with learning activities, such as viewing of video lectures, undertaking of quizzes, posting in discussion forums, and interacting with the courseware is captured.Data captured from MOOCs can provide valuable information for educators by analysing the patterns present in the behaviour of learners.
A common pattern observed is large early dropout and low completion rate of a particular course. Early dropout predictions will provide a framework for developing mechanisms in MOOCs that provide individualized guidance and small-group support,
Increasing retention rates. Thus, an effective way to predict the outcome of a MOOC is needed.
To predict the outcome of user course based on behavioural and demographic patterns.
In this study an extensive data pre-processing and analysis techniques have been performed and exploratory analysis is performed.
The dataset used in this paper was obtained from Harvard University It is divided into these main parts: (i)Dataset -Harvard Dataset
(ii)Data Pre-Processing and Imputation
(iii)Exploratory Data Analysis
After the pre-processing and exploratory data analysis we apply supervised learning models and perform the comparative study of those models. Results and Analysis of these algorithms is performed.
Downcasting of integer and float values. The objects are converted into categorical values. Almost by 82 % reduction in memory usage is obtained and thus further preprocessing can be done. FInal Memory reduced from 381MB to 69MB .
Deleting rows. Replace with mean/median/mode. Assigning unique category. Predicting the missing values using supervised machine learning algorithms. Using algorithms which support missing values.
The correlation heatmap was applied to measure the dependency between the behavioral data and learners certification.So the certification is highly positive correlation with ndaysact which coefficient value 0.68 and with no of events and no of chapters its coefficient value is 0.64 so it is moderately positive correlation. The plot of correlation heatmap, which indicates a positive relationship between three behavioral attributes and the target variable. So the attributes that are highly correlated can be used to reduce the dimensionality of the dataset. Only one of the two attributes can be taken into the final classification model.
Applying Supervised ML Algorithms: We are going to use the supervised machine learning algorithms for classification problem 1.Random Forest.
2.Decision Tree.
3.Gaussian Naive Bayes.
4.Gradient Boosting.
5.Linear Discriminant Analysis.
6.Logistic Regression.
7.Ada Boosting.
8.Multi Layer Perceptron(Neural Network).
Comparison of Algorithms :
[1] A. D. Ho et al., “HarvardX and MITx: The First Year of Open Online Courses, Fall 2012-Summer 2013,” SSRN Electron. J., no. 1, pp. 1–33, 2014.
[2] A. D. Ho et al., “HarvardX and MITxௗ: Two Years of Open Online Courses Fall 2012-Summer 2014,” SSRN Electron. J., no. 10, pp. 1–37, 2015.
[3] Raghad AL-Shabandar,Abir Hussain,Andy Laws Robert Keight, Janet Lunn Naeem Radi " Machine Learning Approaches to Predict Learning Outcomes in Massive Open Online Courses", 2017