Patient Stay Duration Predictor

• 

  Feature importance determination using XGBoost.

Inspiration

When pandemic strikes and people are being constantly admitted to hospitals for treatment, it is important to know how long a patient might stay for proper resource allocation. With limited space in hospitals and constant urge to develop a framework for proper planning, we have developed a prototype GUI which estimates the period for which a patient might be admitted based on certain simple attributes.

What it does

The prototype simulator is capable of simulating how long should a patient be ideally admitted to a particular hospital. Upon entry of ~20 simple features, the simulator produces the estimated time the hospital can expect the patient to be admitted. This can be linked to any resource planning interface the hospital may be using.

How we built it

Built entire using Python 3.6, the GUI was built using the tkinter library. For the generated models, sklearn was used for Logistic Regression, Random Forest, and XGBoost algorithms. The neural network was built with Pytorch. Feature importance was also determined using XGBoost.

Challenges we ran into

The model needs to be tuned further for accurate predictions and to prevent any false predictions. The primary challenge we faced was to improve the accuracy of the model within one night of presentations.

Accomplishments that we're proud of

We were able to develop a fully functional prototype within one night's work. Although it can be improved further, the simulator is up and running, using different models in the backend. This can be commercialized post further improvement!

What we learned

We learned the importance of resource allocation. And also how to integrate deep learning models with the python package tkinter, to create an easy to use user interface.

Built With

• kaggle
• python
• pytorch
• tensorflow
• tkinter