FRMAdam-iTransformer KAN: A Fractional Order RMS Momentum Adam Optimized iTransformer with KAN for EEG and ECG Prediction

Accurately predicting electrocardiogram (ECG) and electroencephalogram (EEG) signals holds significant clinical importance. Although the iTransformer model performs well in long-sequence predictions, it still faces challenges such as low computational efficiency and gradient explosion. To address these issues, this paper proposes a fractional order root mean square momentum adaptive moment estimation optimizer iTransformer model based on Kolmogorov Arnold (FRMAdam-iTransformer KAN). This model incorporates the Fractional Root Mean Square Momentum Adaptive Moment Estimation optimizer (FRMAdam) and replaces the fully connected output layer of the original iTransformer with a KAN structure to improve prediction accuracy. Experimental results on ECG and EEG datasets demonstrate that the FRMAdam-iTransformer KAN model achieves a Mean Squared Error (MSE) of 0.0004 and a Mean Absolute Error (MAE) of 0.0137, representing reductions of 76.47% and 59.59%, respectively, compared to the original iTransformer model.