Improved Model for Stroke Prediction on Wearable Devices with CNN_LSTM_KAN

In order to solve the problem of accuracy and timeliness of stroke prediction on wearable devices, we proposed an improved model. This model combines advantages of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and Kolmogorov-Arnold Network (KAN). We use CNN to extract features, process time series through LSTM and predict by KAN, thereby constructing an efficient stroke prediction model. Experiment results shows the accuracy of 0.92, demonstrating the model’s excellent performance in stroke prediction tasks on wearable devices.