Time series data, prevalent in diverse domains, reflects underlying dynamic processes crucial for informed decision-making. Our research marks a modest stride in comprehending these dynamics. In the context of disease surveillance, Time series forecasting and early warning indicators allow us to anticipate plausible infection waves and their severity. While Deep Learning techniques might deliver accuracy, they often obscure underlying dynamics. On the other hand, statistical methods can be either too approximate or overly complex to capture real-world nuances. Our work bridges this gap, harnessing the strengths of both worlds: we derive understanding from the dynamical system from observed data and proceed to forecast based on this knowledge. This fusion of understanding and prediction will be of paramount interest to decision-makers across various domains. To this end, we propose a data-driven ‘Higher Order Dynamic Mode Decomposition’ (HODMD) based time-series forecasting that decomposes complex time-series data into coherent modes and provides accurate forecasts of future trends. By leveraging the capabilities of HODMD, we aim to enhance time-series forecasting and early warning detection in the context of disease surveillance, which holds significant potential for policymakers and healthcare officials to make timely interventions. The COVID-19 India data has been used to validate the potential of our approach. Contribution of this work is two fold. The proposed approach not only accurately predicts infection waves well in advance but also maintains competitive performance metrics. This makes our method a valuable tool for decision-makers and healthcare officials, offering insights that can guide strategic planning and interventions effectively.