(114a) The Data Analytics Triangle

When analyzing real-world data sets, there is no one-size-fits-all algorithm. This talk discusses the importance of and methods for interrogating a dataset to understand its properties and better select appropriate data analytics methods. The presentation uses the framework of the “Data Analytics Triangle,” which organizes datasets by associating each of the vertices of the triangle with one of three characteristics: nonlinearity, correlation, and dynamics. The edges of the triangle represent datasets that have two of these characteristics, whereas the center represents datasets that have all three characteristics. We then associate each of these parts of the triangle with different data analytics methods, so that the data integration and data analytics triangle guide the user to the best data analytics methods to apply for a particular data set. The final algorithm to apply then depends both and the properties of the dataset and the goal of the data-based model (e.g., prediction, classification).