This repository has all the required files for building an ML model to predict the severity of acute ischemic strokes (brain strokes) observed in patients over a period of 6 months.
View the Project on GitHub ritvik-chebolu/Acute-Ischemic-Stroke-Prediction
Using a machine learning based approach to predict hemorrhagic stroke severity in susceptible patients.
Doctors could make the best use of this approach to decide and act upon accordingly for patients with high risk would require different treatment and medication since the time of admission. This could save lives since a time of action plays a crucial role in determining the lifespan of a patient stuck in coma. In return, such an approach where a model is trained over time with more such datapoints (new patients) could save more lives thereby increasing the medical standards.
The primary assumption made was that all medical factors could be contributing factors to predict the severity of strokes in patients. Spoiler alert, turns out, they do.
The goal of using an Ensemble Machine Learning model is to improve the performance of the model by combining the predictive powers of multiple models, which can reduce overfitting and improve the generalizability of the model. Out of the two approaches for an ensemble model, simple/weighted average and the majority vote, the former seemed to be the best way to proceed since it has a weighted contribution to the final prediction. The majority vote on the other hand, only considers the most common prediction among individual models.
With a relatively smaller dataset (although quite big in terms of a healthcare facility), every possible effort to minimize or eliminate overfitting was made, ranging from methods like k-fold cross validation to hyperparameter optimization (using grid search CV) to find the best value for each parameters in a model.
This repo has all the project files for building an ML model to predict the severity of acute ischemic strokes (brain strokes) observed in patients. For quick navigation, use the following links:
© 2021 Dr. Harshika Chebolu
All copyrights of the dataset belong to Dr. Harshika Chebolu, Post Graduate in General Medicine at Gandhi Medical Hospital, Hyderabad, India.
Having trouble understanding or implementing this project? Check out the documentation or create a pull request and I could help you sort it out.