Predictive Modeling of Cerebral Strokes: An ADASYN-RF Approach for Imbalanced Data

Cerebral stroke is a condition that occurs when blood flow to the brain suddenly stops, and the cells in the brain subsequently die due to lack of oxygen and nutrients. A stroke is associated with risk factors that are mainly linked with lifestyles today, including metabolic syndromes such as high "glucose level, heart diseases, obesity, and diabetes".Current study develops a stroke prediction using the machine learning algorithms: "Logistic Regression (LR), Random Forest (RF), and K-Nearest Neighbors (KNN)". The dataset required for the above study was sourced from the Harvard Dataverse Repository. The "clinical, physiological, behavioral, demographic, and historical data" are included in thisdataset. In this respect, the imbalance of classes would be handled by employing over-sampling techniques, including "SMOTE, ADASYN, and ROSE". This paper proposes a new hybrid machine learning model by combining ADASYN with Random Forest, known as ADASYN-RF, where ADASYN will resample the imbalanced dataset, then Random Forestis applied on the resampled data. Besides, other machine learning models and oversampling techniques are employed for the comparison. Surprisingly, the ADASYN-RF model is able to achieve the highest detection accuracy of 99% mentioned herein, proving its efficiency in stroke prediction. This method thus provides an inexpensive and precise tool for clinical diagnosis of stroke.