Prediction of the convective heat transfer coefficient (or Nusselt number) and heat transfer rates in air-cooled systems are crucial for optimizing the performance of thermal systems. Higher rates of heat transfer indicate efficient thermal convection. Accurate predictions of heat transfer rates help determine fluid flow rates, appropriate materials for construction, and other operating conditions for optimal performance of thermal systems (e.g. heat exchangers). This enables the design of energy-efficient systems with improved overall efficiency, which can manage heat effectively at reduced energy consumption rates and operational costs. In addition, the comparison of predicted and actual heat transfer rates enables quantitative evaluation of exergy losses, fouling, or other adverse operational issues. Machine learning (ML) techniques can be employed to assess internal forced convection through ducts by leveraging their ability to analyze complex flow patterns, model non-linear relationships between governing parameters, and predict system behaviour from experimental or simulated data. The present study explores the use of different ML models to predict the Nusselt number and heat transfer rates in two generic configurations of heat exchangers. For the Nusselt number prediction of the first setup, the results indicate that linear regression performs best, achieving an R2 value of 0.996, an RMSE of 0.798, and an MAE of 0.662. For cold fluid outlet temperature (T2) prediction of the second configuration, whose performance yielded a different but similar type of dataset, linear regression once again demonstrated good performance with an R2 value of 0.9967, an RMSE of 0.4913, and an MAE of 0.4002. Similar trends were seen in the prediction of heat transfer rate and the hot fluid outlet temperature. Therefore, ML models present a viable alternative to traditional empirical methods. This integration of ML into thermal system analysis offers new insights into optimizing heat transfer processes and improving efficiency.