Using the LSTM time series model with KAN to forecast futures prices

This study introduces a hybrid model that combines LSTM (Long Short-Term Memory Network) and KAN (Kolmogorov-Arnold Network) to improve time series forecasting accuracy for futures prices. The model structure uses LSTM to capture temporal dependencies and incorporates a Convolutional Neural Network (CNN) for feature extraction. A KAN computational unit is also included to enhance the model’s ability to address nonlinear market dynamics, demonstrating clear advantages in handling complex data structures.Effectiveness is evaluated through a comparative analysis with traditional time series forecasting methods, such as ARIMA (Autoregressive Integrated Moving Average) and LSTM, along with the CNN-LSTM-KAN model. Experimental results indicate that the KAN-enhanced model outperforms in key performance metrics like MSE (mean square error) and MAE (mean absolute error), confirming that KAN integration enhances the model’s ability to handle market changes and unpredictable conditions.