Inspiration
As of 2021, 1 in 9 Singaporeans has diabetes and it is projected that 1 in 3 Singaporeans is at risk of developing diabetes. It is an insidious medical condition often going unmanaged that leads to cardiovascular disease, nerve damage (neuropathy), kidney damage (nephropathy), eye damage (retinopathy), foot damage and Alzheimer’s. Unplanned hospital re-admissions are costly and disruptive to patients’ work schedules, when they are in fact preventable. We were inspired to create an intelligent model to catch and identify these cases.
What it does
Our TabNet model identifies patients with high risk of re-admission so that healthcare practitioners can intervene to reduce their risk factors
How we built it
We referenced a research paper on the TabNet architecture and implemented it in code with inspiration from an online GitHub repository.
Challenges we ran into
It was challenging to understand the research paper on the TabNet architecture and implementing it in code thereafter. We also faced difficulties in cleaning and analysing the dataset.
Accomplishments that we're proud of
Implementing the TabNet model to achieve commendable F1 and AUC scores for prediction.
What we learned
We gained exposure to a novel and unfamiliar deep learning model and further bolstered our competencies working with the TensorFlow library. We also understood the value in leveraging on open-source repositories to implement our ideas instead of trying to reinvent the wheel.
What's next for this TabNet model
Further hyperparameter tuning to increase the generalisability of the model to new unseen data.
Collecting a more updated and balanced dataset for better training of the model.
The model is versatile - it can be trained to predict hospital re-admission for other categories of patients, not just diabetics.
Log in or sign up for Devpost to join the conversation.