A Predictive Modeling to Assess the Underlying Risks of Stroke

Stroke has a significantly detrimental impact on the human body which necessitates rapid medical care and treatment. There have been widespread initiatives to enhance stroke detection and treatment in response to the overall costs associated with stroke. Early identification and appropriate therapy are essential for minimizing risk to the affected region of the brain and preventing secondary problems. A supportive decision-making model has been designed in this study to analyze the electronic health records of the patients. The automated model is supplied with an extensive set of health and lifestyle factors from the patient’s database in order to make the necessary stroke likelihood prediction. This predictive analysis can reveal the major and interconnected health factors which can raise the probability of stroke. A two-phase designing process is employed to construct the automated prediction. In the first phase, different well-known tree-based models namely Decision Tree, Random Forest, Extra Trees, Gradient Boosting, and AdaBoost are implemented with necessary hyper-parameter adjustments. Random forest is the superior model which has shown the best possible outcome in this phase. The comparative analysis drawn among the employed models highlights the enhanced predictive accuracy of 0.9482 as exhibited by the Random Forest model. However, the f1-score of this model is substantially inferior. Hence, in the second phase, a k-fold cross-validation is further applied to this model to improvise the predictive efficiency of the random forest model. Finally, a promising efficiency is delivered by the cross-validation method with an accuracy of 0.95323 and F1-score of 0.96161.