A courtyard is an architectural design element which is often known as microclimate modifiers and is responsible to increase the indoor occupant comfort in traditional architecture. The aim of this study is to conduct a parametric evaluation of courtyard design variants in a residential building of different climates with a focus on indoor thermal comfort and utility costs. A brute-force approach is applied to generate a wide range of design alternatives and the simulation workflow is conducted by Grasshopper together with the environmental plugins Ladybug and Honeybee. The main study objective is the evaluation of the occupant thermal comfort in an air-conditioned residential building, energy load, and cost analysis, derived from different design variables including courtyard geometry, window-to-wall ratio, envelope materials, heating, and cooling set-point dead-bands, and building geographical location. Furthermore, a Deep Learning model is developed using the inputs and outputs of the simulation and analysis to transform the outcomes into the algorithmic and tangible environment feasible for predictive applications. The results suggest that regarding the thermal loads, costs, and indoor thermal comfort index (PMV), there are high correlations between the outdoor weather variation and dead-band ranges, while in extreme climates such as Singapore, courtyard spaces might not be efficient enough as expected. Finally, the highly accurate deep learning model is also developed, delivering superior predictive capabilities for the thermal comfort and utility costs of the courtyard designs.