An Algorithmic Framework For Classification In Metric Space Research And Application

Most existing classification algorithms are for multi-dimensional data, while algorithms for other data types may have other problems. Facing the variety challenge of big data, domain-specific solutions are usually low in cost efficiency. It is of great need to apply traditional classification algorithms for multi-dimensional data to other complex data types.We propose an algorithmic framework to solve this problem which combine metric space and traditional classification algorithms, and apply it to hurricane track data. First, complex data types can first be abstracted into metric space, considering only the distance information between data. Second, based on the pivot space model, data in a metric space without coordinates can be transformed into a pivot space with coordinates by pivot selection. Third, traditional classification algorithms for multi-dimensional data can be applied to data in the pivot space. Finally, this framework is applied to the classification of hurricane track data.Experimental results show that the proposed framework works efficiently for the classification of complex data types. Further, higher accuracy can be achieved under the framework than classifying multi-dimensional data directly for many cases. At last, applying this framework to track data classification achieves accuracy more than 80% on average.