DEVELOPMENT OF AN INTELLIGENT STROKE PREDICTION MODEL USING MACHINE LEARNING TECHNIQUES

ABSTRACT Stroke continues to be one of the leading causes of death and chronic disability in most countries in the world, especially those with low- and middle-income levels as access to early diagnostic tools is limited. Traditional stroke risk assessment models which include clinical scoring systems often fail to account the complexity of interdependence of lifestyle, clinical and demographic determinants. The current study formulated a stroke prediction model that is understandable and interpretable in this regard, and applied machine-learning methods alongside organized patient data. The method used entailed the harvesting and processing of information which was retrieved through publicly-available stroke data, and the variables of age, hypertension, cardiac disease, blood glucose level, Body Mass Index (BMI) and smoking habits were taken into consideration. Several algorithms were trained and then assessed with help of accuracy, precision, recall, and F1-score measures along with the support of Logistic Regression, Support Vector Machine (SVM), and Random Forest. Random Forest model was found to be the best performer and had an accuracy of 99.2% hence illustrating an instance of predictive superiority as compared to other methods. Explainable AI methods were also incorporated to improve the interpretability of making each prediction to clarify how a single risk factor contributed to that prediction. The user-friendly interface built using Streamlit will allow a clinician or a general user to feed data and get prediction feedback in real-time. The system effectively helps in the informed medical decision by identifying the high-risk individuals at the risk of stroke. The researchers find that the combination of machine learning and explainable AI could create an effective, affordable and scalable tool to precisely predict early stroke, especially where the resource-limited health care setting exists. The refinements that can be done in the future include including hospital datasets, using the system as a mobile diagnostic application.