Introduction
First, please tell us something about yourself: What hashtags best describe you?
#Multipotentialite
#Positive
#ExpatLife
#NatureLover
Tell us more about them.
A #Multipotentialite has more than one passion and a range of interests. In my case, I went from pharmacy to biotechnology to data science. In another life, I could have been a writer, an engineer, a craftsperson or an interior designer. I like to always have a #Positive mindset. Even when things may not be going my way, I look for the silver lining and make the best out of the situation. I have been an expat for many years and have lived in three countries. #ExpatLife has definitely shaped me in many ways. It can be hard sometimes, but also very rewarding to experience life from a different perspective. It’s a guaranteed way to stretch your mind. #NatureLover. One of the things I love most about living in Switzerland is the picturesque views and the opportunity to immerse myself in nature. It’s a blessing.
About your job: What do you do at the moment?
I have been fully dedicated to the Master’s programme. Before that, I took a career break, and before that I worked in academia as a teaching and research assistant.
What did you do before and why did you decide to do the Masters?
I became interested in data while working on my Master’s in biotechnology previously. The thesis project mainly involved wet lab work, but eventually it included a step of DNA sequencing where I had to analyse the data by using bioinformatics tools. This fascinated me, and I was tempted to pursue a bioinformatics or data science degree. After much deliberation, I decided on data science, as it is a broader field where I would learn about a wide range of technologies, concepts and data types.
The Project
Please tell us about your research project.
Given my background in healthcare and life sciences, I was keen on choosing a thesis project in health data science. Recently, several research groups have been looking for ways to use the huge data pools collected from wearable devices such as heart rate monitors and step counters and to help people with chronic conditions, for example diabetes, to manage their illness.
Diabetics are at risk of hypoglycemia, i.e. very low blood sugar levels, which can be dangerous, especially when it sets in during sleep. They therefore have to check their blood sugar levels regularly. To date, sugar levels can be measured only with invasive methods, such as finger pricking or a continuous glucose monitor, a device attached to the body that measures sugar levels every five minutes.
From the idea to the machine learning model
Overview of the project design.
The research question is: “Can we use health data collected by wearables (such as heart rate, blood oxygen levels … etc.) to detect hypoglycemia non-invasively?”
Research in this regard predominantly focuses on medical-grade wearables because of their accuracy. However, our project studies the possibility of achieving good results using commercial-grade wearables. If successful, the results would mean a more practical method that puts such wearables in easier reach for end users.
To move ahead with this project, I needed funding and connections, which I luckily received from everyone involved. So, here’s a huge Thank You to all who made this project possible!
Results and Findings
What data and method did you use, and what did you learn or hope to learn?
Once we received the approval from the Ethics Committee, we started recruiting Type 1 diabetic patients. The participants were given two wearable devices. The first was an Apple Watch Series 8 to collect health data on heart rates, heart rate variability, blood oxygen levels, respiratory rates and step counts, among other things. The second was a continuous glucose monitor (CGM) to measure sugar levels every five minutes.
The data was collected over ten days, during which the participants entered information about their mealtimes, medication and physical activity in a diary. After the ten days, we combined the data from the Apple Watches, CGM and the diary, prepared it and used it to train machine learning models to predict hypoglycemia.
The results look promising. Despite the many limitations of using a commercial wearable, we could build models that perform well. We also understood which variables were necessary for the model to achieve a good result.
Day and night data
Models were built separately for data collected during the day and at night. This figure shows the contribution of different feature groups model predictions.
A look at the collected data
Model performance for one of the participants.
How can your insights help society?
This pilot project provides promising insights about integrating algorithms for hypoglycemia warnings into commercial-grade wearables. Managing a chronic disease like diabetes is complex, and thus providing patients with an effective means of managing their condition can make a huge difference.
How would you like to pursue your project in future?
Two things come to mind in connection with substantiating the project results. Firstly, I want to look into using other commercial wearables such as a FitBit to ensure that the results are reproducible and to study this topic on a larger scale. Secondly, I think it’s worth it to conduct a similar study with children, as they unfortunately also suffer from Type 1 diabetes.
How did your studies influence the project?
The idea for the thesis was actually inspired by a project required for one of the modules. During the Master’s programme, I learned about many concepts and tools that I needed for the project, such as Python coding practices, data wrangling methods, different machine learning techniques, AWS services, among others.
What advice would you give others starting on a similar project?
Dare to do something new, work with supportive people and be very patient (I had to extend my thesis submission deadline twice)! Take your time to process the data before proceeding with modelling and watch out for the garbage-in-garbage-out effect!
And finally, what new hashtag are you aiming for?
#Growth both at the personal and the career levels.
We would like to thank Yasmine Mohamed for her dedication and for sharing these valuable insights.
Applied Data Science Professional Portraits
Programme Information | Contact | Info-Events
Join us and visit our info-events and find out more details in a personal conversation with the Head of the Programme.
Please contact us for an individual and personalised advice:
Tel.: +41 41 228 42 53 / E-Mail: master.ids@hslu.ch
Find more information here:
- Course structure and modules
- Admission and registration
- Professional portraits & study insights
- Working and studying
- Generalist profile
- FAQ
