Deep Learning-Based Financial Time Series Forecasting via Sliding Window and Variational Mode Decomposition

Financial time series forecasting relies on historical data and time series modeling to predict key financial indicators, such as stock prices, indexes, returns, and volatility. Accurate forecasting helps identify market trends and volatility, supports national financial regulation, and assists institutional investors in making informed investment decisions. Traditional econometric models include ARCH (Autoregressive Conditional Heteroske-dasticity Model) and GARCH (Generalized-ARCH), which describe volatility clustering and leptokurtosis in financial time series [1][2]. In the 21st century, deep learning has become prominent. Neural network models such as convolutional neural networks (CNN), deep belief networks (DBN), and autoencoders (AE) have been widely applied to sequence prediction. Among these, recurrent neural networks (RNNs) and particularly long short-term memory (LSTM) networks [3], introduced by Hochreiter and Schmidhuber in 1997 [4], address vanishing gradient problems and are suitable for capturing long-term dependencies.