FAN-TSF: A Frequency Adaptive Normalization Approach for Non-stationary Time Series Forecasting on Stock Market Data

Financial time series forecasting remains a focal point of research in finance due to its crucial role in investment decision-making and risk management. However, the highly nonlinear and non-stationary characteristics of financial markets pose significant challenges for prediction. This paper introduces a Frequency-Adaptive Normalized (FAN) time series prediction model that enhances forecasting accuracy through an innovative frequency domain analysis approach. The model employs a dual-path architecture, incorporating frequency-adaptive normalization mechanisms and residual learning strategies, which effectively captures both the periodic patterns of time series and accurately models fine-grained market fluctuations. Experiments conducted on the TSLA stock dataset demonstrate that the FAN model achieves substantial improvements in both predictive accuracy and computational efficiency compared to traditional methods. Notably, the model exhibits robust performance when forecasting during periods of intense volatility. Ablation studies further validate the necessity of each model component, providing new research directions for financial time series prediction.