Predicting multi-disciplinary design performance utilizing automated topic discovery

Increasing the complexity of engineering design projects expands of the diversity of required topic knowledge. Multidisciplinary design processes have the need for expertise from multiple fields of study. In the context of mass collaboration within engineering design, positioning key members within multi-disciplinary teams is of great importance. Determining how each discipline impacts the overall design process requires an understanding of the mapping between competency and performance. This work explores this mapping through the use of predictive models composed of various regression algorithms. Design performance of students working on their capstone design project is analyzed and the relationship between individual competencies is compared against their overall project performance. Each competency and project is represented as a distribution of topic knowledge to produce the performance metrics. Following the automated topic extraction of the textual data, the regression algorithms are applied. Three topic models and five prediction models are compared for their prediction accuracy. From this analysis it was found that representing both input and output variables as a distribution of topics while performing a support vector regression provided the most accurate mapping between ability and performance.