Learning analytics in the midst of a pandemic

When I began my MSc in Business Analytics in October last year, I couldn’t have foreseen how the disruption we’re currently living through would shape my learning experiences.

First Semester – The beginning.

I chose a course with online delivery to allow me to study whilst working full-time. I was excited to be a trailblazer – studying one of only a small number of online MSc programs available at Imperial College.

In December-19, just as the first cases of the as yet unnamed virus were being identified, I enjoyed meeting classmates on campus who’d flown into London from around the globe. So far in our first semester we had only met via Zoom lectures. It was great to start to get to know everyone outside of the virtual classroom. At the end of the week we shook hands goodbye, already looking forward to our next in-person meet up in a few months’ time.

Second Semester – Stay at home.

The week the virus was named, we were starting our Database course. Online lectures began to close out with the now familiar words ‘stay safe everyone’.  As the WHO declared the pandemic in March, all teaching moved online and the UK went into lockdown. Our course’s digital delivery was suddenly a tried and tested blueprint other programs could adopt. With staff teaching from home, our curriculum continued. I used the Our World in Data dataset for my SQL coursework to query the total number of new cases and new deaths in April.

The USA and UK were the countries with the highest numbers in both categories (ok the World was highest, but please someone tell Pres. Trump that ‘lower than the world’ doesn’t count!)

Third Semester – Follow the science.

Models of the disease transmission, including Imperial’s model, have been central to shaping the govt policy response.

Will I be allowed to meet up with a couple of friends any time soon? Will my sister’s wedding go ahead?

It’s May and Week 9 of the Network Analytics module titled ‘Epidemic Models’ may help.  I learnt that the first step in modelling disease transmission is this question:

Is a person immune following infection, and if so how long for?

The answer allows modellers to pick between one of 3 models transmission (SIR, SIRS, SIS) depending on whether or not an infected person can catch the disease again.

Although this looks straightforward, whether reinfection is possible is a huge unknown for a never-before-seen virus. And this is just the first step in modelling transmission. Layering on the impacts of closing businesses, shutting schools, 2m, masks, handwashing in reducing cases are all secondary to this choice of model structure.  No wonder there have been large discrepancies between forecasts and such a wide-ranging debate on the strength of measures that need to be imposed to slow transmission.

Fourth Semester – What’s open?

As summer arrived, universities were still shut. We sat our exams online – 4 exams over 4 days. Sitting back-to-back exams isolated at home, our class bonded via this virtual experience. Although we couldn’t physically walk out of the exam hall and vent about the most difficult questions, we used our WhatsApp group to encourage and support each other. As I walked around the block to unwind after each exam, I was still finding a way to connect with classmates spanning the full spread of time zones.

After exams we were straight into our Analytics in Business module. For our case study we were tasked with recommending targeted incentives to help retain customers at a large UK gym chain. We used historic membership data to build a logistic regression model to predict which customers are most likely to churn.

When leisure centres re-opened in July, I dived into a near empty pool for the first time in months. The huge retention challenge facing gyms was evident.  Our business case study was set in 2016, but our finding that those on annual memberships in highest affluence groups were least likely to give up their memberships seemed very pertinent.

Fifth Semester – First two Electives

In term 5, electives begun and we got to choose our modules.

1. Five item maximum.

Having seen all the supply chain disruptions in recent months, I chose to study Logistics Analytics. I learned that companies tend to re-order only when inventory drops below a predetermined threshold.  Monte Carlo simulation can help set this re-order point to give a high chance that demand before new goods arrive can be met by existing stock and customers won’t go home empty-handed. For products with lots of demand uncertainty and longer replenishment times companies hold higher levels of safety inventory to reduce risk of stockouts.

In Q1 we saw demand for many essential goods spike practically overnight, and travel restrictions shutting down the supply chain. With such a sharp rise in demand and such a sudden slow down in supply, even a model targeting 99.99% probability of no stockout wouldn’t have provided enough safety stock. No wonder toilet roll ran out!

2. Protect the NHS. Save Lives.

Another area that’s been entirely disrupted is routine healthcare. For my second elective I chose to study Healthcare Analytics. I was tasked to build a decision tree model using data from 2014 to predict missed GP appointments. But we’ve had a shutdown of most non-urgent medical appointments for months now. The huge disruption to routine NHS treatment and its knock-on impact on our health is yet to unfold. When we emerge from the pandemic, can analytics help us to lessen chronic consequences through targeted healthcare catchup programs?

Next Semester and Beyond

Heading into 2021, I’ll be studying 3 more electives and undertaking my final project. I’ll also be staying close to the quote:

“All models are wrong but some are useful” – George Box

Hopefully we’ll use our growing understanding of managing the now not-quite-so-new new disease to gradually get back to giving friends and family a hug.

The world around us has shifted. In due time, we’ll seek to mitigate long-term impacts across health/business/education/….. Many forecasts into 2021 and beyond will be necessarily using data from before times to build models that guide decision-making.

I’ve been learning lots of analytical methods since the start of my MSc. But when I’m analysing data, I’ll be staying grounded in the fundamental teachings of my actuarial training: calling out assumptions and limitations of models. And above all, I’ll be predicting a range of outcomes in our uncertain world.

Stay safe everyone – and let’s hope it’s cases down, life up in 2021!

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: