The goal of this research is to create a performant and comprehensible decision support system
for longitudinal big data, making it applicable in critical domains, wherein wrong decisions made
can have significant repercussions, such as electronic healthcare and telecommunications.
In order to create such a platform, an architecture and algorithms will be designed that can (i)
process the vast amount of longitudinal data efficiently, (ii) construct an ensemble of white-box
models using observational data & prior knowledge and (iii) translate this ensemble into a single
interpretable model in order to provide decision support for experts. To create such a platform,
challenges concerning the following topics will be tackled:
1) Incorporating prior knowledge in predictive models, using Semantic Web technologies, to
improve learning speed and the generalization capability, leading to a higher prediction
performance
2) Improving comprehensibility of white-box ensembles by combining the different models into a
single, interpretable model with an enhanced predictive performance compared to the models
individually
3) Efficient extraction of interpretable features with high discriminative power from longitudinal
data in order to be tractable in a big data setting
4) Design of proposed platform and newly realized algorithms compliant with big data frameworks
to ensure high performance and scalability