Enhanced stroke prediction using stacking methodology (ESPESM) in intelligent sensors for aiding preemptive clinical diagnosis of brain stroke

Abstract

According to the World Health Organisation (WHO), brain stroke is the second leading cause of death and disability worldwide in terms of key ICD (International Cause of Death) categories. Every 4 min, on average, a stroke-related mortality occurs in India alone.Machine learning (ML) has developed into a potent tool in healthcare settings, providing individualised therapeutic treatment for stroke patients. ML gives an accurate and speedy prediction conclusion. However, it is clear from the research that better training quality is required, along with the identification of the best classifiers for combining to improve prediction accuracy. In this article, we have suggested a framework to aid diagnosis and Clinical treatment for patients with stroke.In this study, we introduced a system called Enhanced Stroke Prediction Ensemble using Stacking Methodology (ESPESM) in Intelligent Sensors, which expands on the previously successful SPE using Stacking and leverages a hybrid method considering feature engineering and ensemble classification.The fundamental classifiers for the proposed stacking prediction model were Random Forest (RF), K-Nearest Neighbours (KNN), Logistic Regression (LR), Support Vector Machine (SVM), Naive Bayes (NB), Gradient Boosting Classifier (GBC), Decision Tree Classifier, Stochastic Gradient Descent(SGD), and Bernoulli NB(BNB),while Random Forest was selected as the meta learner. Accuracy, Precision, Recall, and F1 score have been used to assess the suggested stacking model's performance.The most accurate models from a pool of potential brain stroke prediction models are selected, and these models are then layered to create an ensemble model. According to prior established literature, stacked ensembles often give a more accurate prediction performance than simple individual models or average ensemble addition. The framework is accomplished by utilising and adapting previously existing methods. They are referred to as “Hybrid Ensemble and Feature Engineering for Stroke Prediction (HEFE-SP)" and “Hybrid Measures Approach for Feature Engineering (HMA-FE)".Additionally, when a dataset is imbalanced, as it is in most medical datasets, the Matthews Correlation Coefficient (MCC) has also been utilised for more accurate evaluation. The accuracy and MCC values of the proposed stacking model, which are 98 % and 94 % respectively, show that it outperforms solo classifiers.