Application of Machine Learning to Analyze the Risk Factors of Stroke
Abstract views: 154 / PDF downloads: 116
Keywords:
Machine Learning, Risk Factors, StrokeAbstract
In 2019, stroke was the second leading cause of death and disability-adjusted life years, globally. 80\% of second strokes have been demonstrated to be preventable by using medication, maintaining a strict diet, and engaging in physical activity. Considering its debilitating effects, early detection of stroke is an important area of interest. Thus, this study aims to identify key risk factors for stroke, to encourage proper monitoring and lifestyle changes that can prevent stroke onset. To determine the most significant risk factors, a machine learning (ML) based artificial neural network model was derived from the Keras library in Python. With an accuracy of 92\%, the model was then applied to different combinations of risk factors using the SelectKBest function. At first, a feature selection strategy using chi-squared scoring was used to select the K best features from a combination of risk factors. These prominent features were then used to train the ANN to predict presence of stroke. The accuracy of the trained model was presented in terms of area under receiver operating characteristic curve (AUC). Average glucose level, age, and BMI were determined to be the most predictive risk factors of stroke. ROC analysis yielded an AUC value of 0.73, which indicates good test performance of the model's determination of the aforementioned most significant combination of risk factors. In addition to confirming the significance of frequently reported risk factors in the existing literature such as average glucose level, age, BMI, smoking status, and hypertension, the model identifies occupation as the next most predictive risk factor for stroke, surpassing even heart disease. Thus, with information on patients, preventative measures can be given based on previously unidentified risk factors like occupation to hopefully avoid the long-term impacts of a potential stroke.
Downloads
Published
How to Cite
Issue
Section
License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.