Gated Recurrent Unit Based on Kernelized Activation for Stock Price Prediction

Stock price prediction has always been a difficult undertaking in the financial market since it not only influencese investment decisions but also has a direct impact on economic advantages.so directly impacts economic benefits. However, stock prices are influenced by numerous factors, making accurate predictions even more difficult. To address this, we propose a time series prediction model based on Kernelized Activation (KAN)—Kernelized Activation Based on Gated Recurrent Unit (KAGRU). The KAGRU model replaces the traditional activation function in the GRU with the spline-based KAN activation function, allowing the model to capture more complex nonlinear characteristics. At the same time, the gated mechanism of GRU is effectively utilized to handle long-term dependencies. Experimental results demonstrate that the proposed KAGRU approach outperforms existing popular time series prediction methods.