Predicting Strokes: A Comparison of Deep Learning and Machine Learning Models

Not being able to quickly detect a stroke. It can lead to serious problems, like permanent brain damage or even death. This is where deep learning (DL) and machine learning (ML) models come in. They can help spot strokes early, making treatments more effective and reducing serious long-term effects. A team in Tehran, Iran, studied 663 patient records from Hazrat Rasool Akram Hospital. These included 401 healthy folks and 262 stroke patients. They used eight well-known models: four from ML (Support Vector Machine (SVM), Extreme Gradient Boosting (XGB), K-Nearest Neighbors (KNN), and Random Forest (RF)) and four from DL (Deep Neural Network (DNN), Feedforward Neural Network (FNN), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN)). Their goal? To see which models were best at predicting strokes. They used methods like 10-fold cross-validation and fine-tuning to avoid overfitting. They also looked at how easy the models were to understand using something called Shapley Additive Explanations (SHAP).They checked the models' performance based on accuracy, specificity (how well the models identify healthy people), sensitivity (how well they spot stroke patients), F1-score (a balance of accuracy and sensitivity), and ROC curve metrics. Among the DL models, LSTM was great at spotting strokes (96. 15% sensitivity). FNN was best at everything else (96. 0% specificity, accuracy, and F1-score, with a 98. 0% ROC score). For ML models, RF was the champ with 99. 9% sensitivity, 99. 0% accuracy, 100% specificity, 99. 0% F1-score, and a 99. 9% ROC score. Overall, RF was the top dog. DL models generally did better than ML, except for RF. They had sensitivities from 93. 0% to 96. 15%, specificities from 80. 0% to 96. 0%, accuracies from 92. 0% to 96. 0%, F1-scores from 87. 34% to 95. 0%, and ROC scores from 95. 0% to 98. 0%. ML models had wider ranges, from 29. 0% to 94. 0% sensitivity, 89. 47% to 96. 0% specificity, 71. 0% to 95. 0% accuracy, 44. 0% to 95. 0% F1-score, and 64. 0% to 95. 0% ROC score. This study shows both DL and ML can predict strokes well. But RF models were the best, showing that combining these technologies could really help in early stroke detection, saving people from terrible consequences.