Bridging the Gap: Enhancing clinical accuracy in stroke prediction using a hybrid RNN-Random Forest Model

Stroke stands as one of the major causes of morbidity and mortality across the world, and timely and precise prediction of the condition is essential, especially when it comes to successful management of the problem. But conventional diagnostic tools usually perform poorly on high-dimensional data with class imbalance characteristic of medical data. It is a study of a comparative analysis involving a detailed discussion of Machine Learning (ML), Deep Learning (DL), and hybrid ensemble to support stroke risk prediction. The suggested approach establishes a powerful data processing flow, which implies Application of Random Oversampling to overcome the imbalance in classes and Principal Component Analysis (PCA) to extract features successfully. We have tested twelve different classifiers, which include traditional algorithms (Random Forest, SVM and XG Boost) and deep neural networks (ANN, CNN, and RNN). Additionally, we proposed new hybrid models (ANN-RF, CNN-RF and RNN-RF) that aim to combine the power of deep learning in extracting the features with the power of Random Forest in classification. The experiment shows that hybrid models are better predictors. Standalone Deep learning models such as CNN and RNN had 95.94% and 95.78% accuracy respectively but the hybrid models still excelled over them. RNN-RF (Recurrent Neural Network with Random Forest) model has got the best accuracy of 96.86, which is better than 96.81 of the standalone Random Forest as well as other hybrid models. This evidence suggests that the combination of sequential pattern recognition of RNNs and ensemble decision-making of Random Forests can enhance diagnostic accuracy substantially, which is an excellent framework that can be used in clinical decision support.