The article discusses the problem of reconstructing the heat transfer coefficient under fourth-kind boundary conditions using swarm algorithms, considering the kappa parameter's variation in three value selection ranges. The analysis was conducted based on the time-varying nature of this coefficient, which represents the physics of the heat exchange process at the interface between the casting and the mold. A three-tier division was adopted: low (0-400 W/m2K), medium (400-900 W/m2K), and high (900-1500 W/m2K) kappa values, with appropriately selected reference values (250 W/m2K, 500 W/m2K, 1000 W/m2K). The geometry of the problem was represented by a model with 576 finite elements, which allowed for a detailed analysis of heat behavior in the casting form and the casting itself. To solve the inverse problem, two metaheuristic optimization algorithms, Artificial Bee Colony (ABC) and Ant Colony Optimization (ACO), were applied and implemented in a dedicated computational environment. The verification of the correctness of the results involved comparing the determined coefficient values with their reference counterparts. A functional based on the L2 norm was adopted as the assessment criterion, measuring the difference between the computed and reference data. Numerical experiments were conducted for various configurations with different numbers of individuals, iterations, and levels of input data disturbances. For each case, three independent runs were performed to assess the reproducibility and stability of the results. Particular attention was given to how both algorithms behave depending on the range of kappa values and the presence of disturbances. The analysis of the results indicates significant differences in the selection of the thermal conductivity coefficient for individual methods, disturbances, and how the parameter is mapped depending on the range of its values. Both approaches exhibited varying stability across different test ranges and differing sensitivity to input data variability. It was observed that the characteristics of the kappa interval and the type of algorithm used directly impact the dispersion of results and the quality of the fit.