One of the challenges in executing an MDO system is the selection of the best performing MDO architecture in terms of computational effort for a given MDO problem. To address this, several comparative studies have been created, which analyze the relative performance of MDO architectures with respect to internal features of MDO problems, such as the number of design variables and nature of interdisciplinary coupling. However, the existing research is limited in applicability. An inclusive prediction model, that can recommend the best MDO architecture for any MDO problem is not found yet in literature. In this thesis, the intention is to conduct a thorough comparison of two commonly used MDO architectures with respect to a database of sequentially generated MDO problems, having a range of internal features. Following this, a prediction model of MDO architecture is to be developed by applying a suitable machine learning algorithm on the generated database.