Massive Open Online Courses Planning
Massive open online courses (MOOCs) have been growing tremendously in the past five year, meeting the demand of making learning more flexible and accessible. MOOCs differ from traditional online courses because they offer interaction and feedback to support the students throughout the course. Many institutions have jumped on the MOOCs band wagon via online platforms, and numerous courses in topics ranging from computer science, business and the creative arts have been posted.
ClassCentral.com is a website that provides content curation for MOOCs. Class Central lists information which include start times of active sessions, duration, institution background, online platform links, summary of course, and reviews for each course. In browsing the website, students are able to make better informed online learning choices.
While MOOCs offer ample flexibility and accessibility, there are also concerns of effectiveness and productivity. Questions that have risen include:
- How many courses can a student take given a specific time frame, if the goal of the student is to obtain certifications?
- Given a student's specific preferences, what are the best ranked courses?
- How to characterize course duration for courses that have flexible deadlines?
Data relevant to 14,000+ online courses scraped from classcentral.com have been utilized to answer the questions above. Meta data and exploratory data analysis visualizations:
How many courses can a student take given a specific time frame, if the goal of the student were to obtain certifications? If the student's goal is to complete the courses, the subset of courses we wish to study is the group that offers certifications upon completion. For this group, we observe that the average course duration is around 100 hours, over the span of 8 weeks. So if a student were to work full time or part time, it's advantageous for students to take one course at a time.
Given a student's specific preferences, what are the best ranked courses? The entire catalog courses is first filtered by preferences that include: language, with or without certification, recently starting, and subject name. Once the sub population of courses is designated, a ranking system of using Bayesian Average Ratings is developed. This algorithm takes into account the number of reviews and ratings data from each course. For example, top three ranked course selection given the preference of Computer Science subject, taught in English, with certificate, recently started or starting soon are:
- Machine Learning, Stanford University via Coursera
- Build a Modern Computer from First Principles: From Nand to Tetris, Hebrew University of Jerusalem via Coursera
- Transport Systems: Global Issues and Future Innovations, University of Leeds via FutureLearn
How to characterize course duration for courses that have flexible deadlines? While course duration may seem to be an easy variable to interpret, there are actually multiple contributors to its value. Courses may be considered as a standalone physical entity, which consists of lecture playback time, homework completion time and exam completion time. These values may be described using normal distributions with bounds around an average. However in reality, course duration is not so much a physical feature of each course, considering interruptions to leture video playbacks and assignment completion delays due to learning inconsistencies. Course duration may have distributions without bounds or patterns because it's determined by each individual user. Therefore feature engineering is utilized separate course duration into three additional features: lecture playback time, flexible deadlines yes, and recommended assignment completion time, all of which have to scrapped from the host platform, such as Coursera. These features are a better indication course duration, and if modeling were to be done on the dataset, recommended assignment completion time could be a candidate predictor.