The new distance measure is an improvement over the current standard whenever a hierarchical arrangement of categorical values is available.

Established process models for knowledge discovery find the domain-expert in a customer-like and supervising role. In the field of biomedical research, it is necessary to move the domain-experts into the center of this process with far-reaching consequences for both their research output and the process itself. In this paper, we revise the established process models for knowledge discovery and propose a new process model for domain-expert-driven interactive knowledge discovery. Furthermore, we present a research infrastructure which is adapted to this new process model and demonstrate how the domain-expert can be deeply integrated even into the highly complex data-mining process and data-exploration tasks. We evaluated this approach in the medical domain for the case of cerebral aneurysms research.

No abstract

Abstract. Biomedical research requires deep domain expertise to perform analyses of complex data sets, assisted by mathematical expertise provided by data scientists who design and develop sophisticated methods and tools. Such methods and tools not only require preprocessing of the data, but most of all a meaningful input selection. Usually, data scientists do not have sufficient background knowledge about the origin of the data and the biomedical problems to be solved, consequently a doctor-in-the-loop can be of great help here. In this paper we revise the viability of integrating an analysis guided visualization component in an ontology-guided data infrastructure, exemplified by the principal component analysis. We evaluated this approach by examining the potential for intelligent support of medical experts on the case of cerebral aneurysms research.

Ontologies are useful for modeling domains and can be used to capture expert knowledge about a system. Genetic programming can be used to identify statistical relationships or models from data. Combining expert knowledge as well as statistical rules identified solely from data is necessary in application domains where data is scarce and a large body of expert knowledge exists. We therefore study if the performance of genetic programming can be improved by incorporating prior knowledge from an ontology. In particular, we include prior knowledge as additional features for genetic programming. The approach is tested with six benchmark data sets where we compare the required computational effort that is necessary to find an acceptable model with and without additional features. The results show that additional features gathered from an ontology improve the performance of tree-based GP. The probability to find acceptable solutions with a fixed computational budget is increased. For noisy data sets we observed the same effect as for the data sets without noise.