Inspiration
Our work is inspired by our heroes situated in hospitals, who work tirelessly to diagnose patients. We hope that through machine learning, simple diagnosis can be automated, allowing for more efficiency in diagnosis of patients
What it does
Our prototype is a machine learning model that is able to relieve the heavy duties of nurses by predicting the seriousness of a patient, allowing nurses to prioritise the patients who are in need of help. The model takes in details such as past medical history of patients and current medical conditions. It outputs the prediction if a patient would survive based on their current conditions.
How we built it
We built our models on AWS SageMaker, accessing our dataset through S3 bucket.
For the processing of our dataset, we normalised data with Box Cox Transformation and Feature Scaling. We dealt with sample imbalance using the imblearn library.
We leveraged on classification ML/AI models such as XGBoost, LightGBM, Random Forest and ANN to predict our results.
Challenges we ran into
Within a short time, we had to implement ML models and carry out exploratory data analysis.
Accomplishments that we're proud of
Our model gave a high accuracy of over 80%. We are also proud to complete this within a span of 1 day. In addition, learning how to integrate AWS Sagemaker is not easy but we are proud to have implemented it.
What we learned
We managed to implement many advanced models in a short amount of time and overcame challenges present in processing the dataset.
What's next for Training Set 01
We hope that hospital systems are able to utilise this to reduce their workload and allow them to focus on other priorities such as patient care.
With an increase in data available, we would hope to optimise our model and improve on accuracies and F1 scoring.
Built With
- amazon-web-services
- ann
- keras
- lightgbm
- python
- sagemaker
- scikit-learn
Log in or sign up for Devpost to join the conversation.