You can't read anything about technology trends these days without reading about Big Data and the power of algorithms. It pops up in education with lots of discussions of education analytics/learning analytics and a pile of other acronyms. I think that the discussion is so intense in education because it’s one of the key sectors that could tap into the power of data to improve business processes – whether that’s improving administration or improving teaching and learning. And it links directly to work our teams are doing with analytics and cloud services. I’m going to share ideas for using Azure Machine Learning in education that will help illustrate what’s possible.
The education sector is awash with data, although it’s often locked away, and it is also full of powerful cases for using algorithms to improve learning and administration. But one challenge is that the skills you need to analyse and use education data are exactly the same skills that are in demand in the rest of the business world – whether that’s because a bank wants to use algorithms to reduce credit card fraud or make more profit trading shares; or a retailer wants to use algorithms to recommend the next product you should buy from their website; or an advertiser wants to put exactly the right advert into your eyeline at exactly the right time. So if you want to use Azure Machine Learning in education you’re going to be competing for the experts with banks, retailers and marketing companies!
There are two ways to solve this problem:
- Data scientists build and share education algorithms that make it easier to analyse data to produce answers
- Make it easier for every day users to be able to apply their expertise to analyse their own data
In this blog post, I’m going to cover the first way – building and sharing algorithms – and then next week, I’ll look at what we’re doing to make it easier for every day users to be able to use intelligent analytics and machine learning.
Part One: Building and sharing algorithms
One of the services in the Microsoft Azure cloud is Machine Learning – a way of using the power of a cloud data centre to do complex analyses without having to build your own room of whirring high performance servers to crunch numbers.
Machine Learning allows you to build the algorithms (“based on these twenty things, it looks like this is going to happen”) and then run them to interpret your own data. Some examples of how this is used today include:
- Estimating demand for a service/product
eg forecasting how many ice creams will get sold next week - Turn speech into text
eg creating captions for TV programmes - Tell you what a picture is
eg identifying an animal in a photo - Identify a person in a photograph, or generic information
eg tell you whether they are male or female, or their age - Recommend products based on what you’ve just bought
eg customers that bought this game, also bought… - Identify which customers are likely to change suppliers
eg who’s going to change phone company - Detect anomalies in data
eg creating an alert when somebody logs on to their social media account from the other side of the globe
In the past, that would have been a massive task, with massive teams of very highly specialised experts and loads of technology – and a long time between having the idea and getting a working system.
But today it’s like many other IT projects – it’s quicker and easier to just get on and build something as a prototype, than to get people together to sit in a room and decide what should be done. And then, of course, you just keep improving what you built in iterations.
And that’s the power of algorithms and machine learning – you keep improving the algorithm as you go along, and through machine learning, so does the system. You don’t need to work out all the rules in advance, but learn as you go.
There is a marketplace emerging for these algorithms – the Azure Machine Learning marketplace has a growing bank of them that include many of the scenarios above, as well as standard statistical models (see the Machine Learning API projects here). Some of these are being created as research projects, others are being created by businesses who will license or sell them to other organisations (eg advanced product recommendation or customer churn algorithms have high commercial value).
So what does this mean for education?
Many of the current experimental projects in other industries have a direct parallel in education. For example, a customer churn prediction algorithm has direct relevance to the student attrition problem in Australian universities (even down to the actual churn rates, where the churn rate for an Australian telco matches the student drop out rate for Australian universities).
Experimental projects are being published constantly – the list today includes over 400 experiments in hundreds of areas :
| Azure ML Experiment | How could that be used in education? | More detailed information |
| Social media sentiment analysis | What are people saying about my university/school? | |
| Movie recommendation | What supplementary course materials match this lecture recording? | |
| Flight delay prediction | What is the likely lecture room capacity needed to optimise campus use? | |
| Predictive maintenance | What does the facilities team focus on to minimise campus disruption? | |
| Fraud detection | Which students are getting somebody else to submit their assignments? | |
| Student problem solving | Will a student eventually be able to solve the problem, based on their first attempt? | |
| Customer Segmentation | How do we divide our 100,000 prospective student into groups for marketing purposes? | |
| Buyer propensity model | How many of our student applicants are likely to start the course? | |
| Student performance – Mathematics | Predict a student’s performance in future tests |
You can find the full list of experimental projects on the Azure Machine Learning Gallery
How can this be used in education?
Today, there are groups of data scientists and specialists using this technology to build algorithms, and also converting algorithms from previous ways of doing the analysis. Some of that work is happening in universities, and some is happening in specialist suppliers to education.
If you’re a budding data scientist, or have access to a data scientist team, then there’s plenty of information (and training materials) on our Azure Machine Learning Studio website if you want to do-it-yourself.
The alternative is to work with one of our advanced analytics partners. For example, S1 Consulting in Sydney and Neal Analytics in Seattle are collaborating on building an advanced student retention system, using big data analytics to predict which TAFE and university students are likely to need help and interventions to keep them on track. This being launched at a briefing event in August in Brisbane on 10th August. They’ve created, in a few months, the kind of analytics system that previously would have cost an individual university millions of dollars in technology and staff time to develop.
Part Two: Making it easier for every day user to use intelligent analytics and machine learning
For the majority of people reading this blog post, the real problem is knowing how to use the information above! (Congratulations and thanks for sticking with it). It’s interesting, but to make use of it, you’ve got to go and find the person in your organisation, or an external partner, who has the technical skills to use the technology. And your critical input is to help them identify the real problems that the data can help solve, and the value to the organisation from solving them.
Next week, in Part Two, I’ll look at our work to try and solve this part of the challenge – to bring the power of machine learning and advanced analytics without needing to be a rocket scientist/brain surgeon.