BRAIN STROKE PREDICTION

Stroke is a leading cause of mortality and long-term disability worldwide. Timely diagnosis and risk assessment are critical for reducing fatalities and improving patient outcomes. This paper presents an integrated, web-based system for comprehensive brain stroke prediction, detection, and preliminary diagnosis. The system adopts a dual-AI approach: a Random Forest model analyzes structured health data to produce a stroke risk probability and personalized recommendations, while a MobileNetV2-based Convolutional Neural Network (CNN) analyzes CT/MRI images to detect hemorrhagic stroke. Key features include robust input validation to reject non-medical images and Explainable AI via Grad-CAM heatmaps for visual interpretability. Implementation employs Flask for the web interface, scikit-learn for tabular models, and TensorFlow/Keras for deep learning. The Random Forest achieves over 95% test accuracy and the CNN achieves approximately 97.5% validation accuracy on the hemorrhage dataset. The platform is presented as a decision-support tool for clinicians with necessary clinical disclaimers..