Orthogonal decision trees for resource-constrained physiological data stream monitoring using mobile devices

This paper considers the problem of monitoring physiological data streams obtained from resource-constrained wearable sensing devices for pervasive health-care management. It considers Orthogonal decision trees (ODTs) that offer an effective way to construct a redundancy-free, accurate, and meaningful representation of large decision-tree-ensembles often created by popular techniques such as Bagging, Boosting, Random Forests and many distributed and data stream mining algorithms. ODTs are functionally orthogonal to each other and they correspond to the principal components of the underlying function space. This paper offers experimental results to document the performance of ODTs on grounds of accuracy, model complexity, and resource consumption.