Decision Tree on Neo4j

Schema

About

Graph databases enable efficient storage of heterogeneous, highly-interlinked data, such as clinical data. Subsequently, researchers can extract relevant features from these datasets and apply machine learning for diagnosis, bio-marker discovery, or understanding pathogenesis. To facilitate this process by saving time for extracting data from the graph database, we developed DTP (Decision Tree Plug-in), containing 19 procedures to assist and generate decision tree algorithms directly on the graph database Neo4j. As a result, our approach required between 0.059 - 0.099 seconds for creating a decision tree of three clinical test data sets, while the csv-based Java decision tree needed 0.085 - 0.112 seconds. Furthermore, our approach proved to be the fastest across baselines in R and Python using csv files instead of nodes as input. A bonus feature being that DTP persists the decision tree in Neo4j allowing for direct visual analysis. Overall, our work shows that integrating machine learning into graph databases saves time and should be applied to a variety of use cases including clinical environments.