The modular housing has increasingly gained the popularity among communities in views of its fast construction and minimum site preparation. The prefabrication technology makes it possible for the easy storage and mass logistics to construction sites. At variance with typical frames, the modular house composes a series of prefabricated lightweight steel panels that are assembled through the precisely designed connections (nuts and bolts and/or welding). The total cost and assembly time are proportional to the number of connections predefined. This paper, therefore, proposes a so-called Comprehensive Learning Particle Swarm Optimization (CLPSO) method to determine the minimum placement of connections necessarily required for the assembly of semi-detached modular houses under applied external forces. The limit state design criteria comply with AISC-LRFD (2016) specifications. The connections adopt the combination of five specially designed nuts-and-bolts patterns, where their possible locations are predefined. Their behaviors are described by the compatibility conditions of displacements at some specific degrees of freedom (in 3D space) associated with interface nodes of steel panels. The proposed CLPSO approach assigns the binary variables to all connection locations and efficiently determines their optimal placement leading to the minimum construction cost. The applications of the proposed CLPSO method are illustrated through the realistic design of public residential houses managed by National Housing Authority of Thailand.