This study aims to develop an intelligent and integrated framework for the predictive monitoring of the performance of air-cooled heat exchangers in refinery applications. The main innovation lies in the simultaneous and systematic investigation of the impact of fluid molecular structure (focusing on hydrocarbon branching) and dynamic operating parameters within a data-driven approach. To this end, a real heat exchanger was configured using simulation software and technical data. Ten hydrocarbon compounds with linear, two-branched, and three-branched structures were selected, and for each compound, one hundred operational scenarios were executed by varying key parameters, including inlet temperature, inlet flow rate, and fouling coefficient. Quantitative findings revealed that altering the molecular structure to longer branched forms significantly increased pressure drop by 2.25 to 3 times under a constant fouling coefficient, resulting in a decrease in heat exchanger performance. This indicates that changes in fluid structure (e.g., branching) have a substantial impact on exchanger performance. Additionally, sensitivity analysis using the Sobol method identified the inlet flow rate as the most influential parameter on exchanger performance, with an index of 0.8086. Accordingly, the operational output of the research is a predictive monitoring framework based on machine learning models, which, by defining three levels of performance indicators (optimal, requiring monitoring, and critical), facilitates early detection of fouling and optimal maintenance planning. The implementation of this system could enhance the reliability of heat exchangers and achieve significant cost savings in operational expenses by reducing unplanned shutdowns.