Knowledge graph embedding for experimental uncertainty estimation

Purpose Experiments are the backbone of development process data-driven predictive models for scientific applications. The quality experiments directly impacts model performance. Uncertainty inherently affects experiment measurements and is often missing in available data sets due to its estimation cost. For similar reasons, very few compared other sources. Discarding based on uncertainty values would preclude models. Data profiling techniques fundamental assess quality, but some dimensions challenging evaluate without knowing uncertainty. In this context, paper aims predict experiments. Design/methodology/approach This work presents a methodology forecast experiments’ uncertainty, given set ontological description. approach knowledge graph embeddings leverages task link prediction over representation database. validity first tested multiple conditions using synthetic then applied large chemical kinetic domain as case study. Findings analysis results different test scenarios suggest that embedding can be used when there hidden relationship between metadata values. also resilient random noise relationship. outperforms baseline if depends upon metadata. Originality/value employment experimental novel alternative current more costly literature. Such contribution permits better repositories improves