The increasing use of time series data across various domains necessitates advanced predictive models capable of addressing nonlinearity and non-stationarity with high accuracy and efficiency. This paper proposes EDM-2DSL, a novel framework that combines Empirical Dynamic Modeling (EDM) with Decomposition of Signal Dynamical System Learning (2DSL) to enhance time series forecasting. By integrating these approaches, EDM-2DSL effectively captures the inherent complexities of dynamic systems, improving prediction accuracy. The framework’s performance is evaluated on datasets from finance and climate change, benchmarking against traditional neural networks such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. Results demonstrate that EDM-2DSL outperforms these methods, offering superior predictive capabilities and robust analysis of complex time series data. These findings highlight EDM-2DSL’s potential as a valuable tool for researchers and practitioners in domains requiring accurate forecasting of intricate dynamic behaviors.