Want to Stop Rebuilding "Expensive" Parts of your Flow? Explicit Builds are the Answer!READ MORE

Atlantic Technological University - Research on Application of Machine Learning to Optimize Learning Analytics & Improve Student Retention

Name:

Ikechukwu Ogbuchi
Dr. Etain Kiely
Dr. Cormac Quigley
Donal McGinty

Country: Ireland {Republic}

Organization: Atlantic Technological University

ATU is a multi-campus technological university in the west and northwest of Ireland that delivers a rich combination of academic and research excellence, quality of life, and opportunity.

Awards Categories:

  • Data Science for Good
  • Responsible AI
  • Excellence in Research

 

Business Challenge:

My research is about the Application of Machine Learning Algorithms to Optimize Learning Analytics. This stems from a Learning analytics project which began in 2015 at GMIT (now officially known as Atlantic Technological University, ATU).

The project started with data analysis expressions using Excel spreadsheets. The goals of the project included gaining deeper insights into various student groups, identifying on time which students were at risk of dropping out, and developing the best ways of improving engagement for our students.

We made progress, but there were lots of data captured from our learning platform which we were not able to use and analyze. More challenges with older techniques on big data created the need to explore machine learning, and this was where I became a part of this project as a research student in 2021.

My initial goals working on the research team were to build, explore the power of predictive models on the data we currently have, and uncover useful insights which could be of value to the project. It was hoped that uncovering these insights would contribute to creating the best learning experiences for the over 20,000 students who are part of the new ATU. ATU was formed from three institutes of technology here in Ireland: Galway-Mayo Institute of Technology, Letterkenny Institute of Technology, and Institute of Technology Sligo.

As part of the team, I made use of the Python Programming Language coupled with Jupyter notebooks to build predictive models and analyze available data for the project. However, at some point, I saw the need to explore other available tools on the cloud. I was looking for a platform I could quickly run machine learning experiments, compare results, and overcome challenges running these models on my device. It was then I started playing around with Dataiku.

 

testATU.PNG

 

Business Solution:

I heard about Dataiku for the first time in a Microsoft Ignite program in March 2021. I was fascinated by their value proposition, so I registered to attend one of their partner sessions on Delivering enterprise AI at scale. I liked the session and subscribed to receive information, but I never really thought about experimenting with the platform until the need came up this year. I was able to complete the Core Designer and ML Practitioner certifications on the Dataiku Academy in June 2022, and they were enough for me to get going.

I found Dataiku very easy for me to use in experimenting with machine learning models. I was able to quickly run different machine learning models for clustering like the K-means, Agglomerative clustering, Gaussian Mixture, DBScan, and lots more with so much ease and less code in Python! This accelerated my ability to see what models were faring best for the project and available in an intuitive user interface. The cool thing here for me was that Dataiku had the option of exporting these models in a Jupyter Notebook if I wanted to make any changes in Python.

The steps followed in this project were mainly data collection, data preparation, data visualization, machine learning modeling, and reporting on results. And the following tools have been used together with Dataiku: Excel, Python Programming, Jupyter Notebook, and Powerbi(for visualization and reporting). We have also devised easy ways to pull data using our developed APIs from our learner management platform and get that data in the right format for analysis.

So far, we are deriving new insights from patterns associated with different student groups, and we are better understanding how to support them best. Feedback from our pilot student groups shows that they are happy about the research and steps we are taking.

Here is a snippet of our current dashboard:

Dashboard ATU.png

And here is a snippet of what doing this in Python looked like:

Python snippet ATU.png

 

 

Value Generated:

So far, we can identify students who are at risk based on their patterns of interaction with their learning systems and have programs in place to support these identified student groups. These intervention programs range from student mentor support, additional classes, financial aid, counseling programs, and lots more. This has created a good sense of belonging for the pilot students that have been identified so far.

We took a survey towards the end of the year, and 97% of the students in our pilot program agreed that they have found the steps we are taking very helpful. One student mentioned to me that from the feedback she felt that the lecturer cares about her genuinely, not just about her grades. That is how we want every student to feel. A college where every kind of student feels like they belong and that we can understand them better.

We have made different conference presentations for the project and have journal papers currently under consideration for publication.

Here is a snippet of some student Feedback results:

Feedback students.PNG

 

Value Brought by Dataiku:

Dataiku has brought value in terms of ease of experimenting with these models. It has saved me loads of useful time. And we all know time seems scarce a commodity these days.

I must say that the one-year free access for academics is impressive. Dataiku platform has great value and making this platform free for research students and lecturers is highly commendable.

I have also found the Dataiku Community helpful. During the first few weeks when I was exploring Dataiku, I found a data scientist, Christy, in the community who asked questions related to the kind of research I am doing. I sent her a message asking to have a meeting, and although it took a few days before she responded (as she was out for a conference), she was very willing to collaborate with me, and we have been having meetings together to learn more about Dataiku usage ever since. I think having such a community where you find such helpful people is important for growth and advancement.

Here are images from use of Dataiku with results:

Models_Dataiku_Atu.png

And here is a quick visual of selected features:

kmeans.png

Share:
Version history
Last update:
‎08-25-2022 06:17 PM
Updated by:
Contributors