Preserving prediction accuracy on incomplete data streams
Authors
O. Parisot, Y. Didry, T. Tamisier, and B. Otjacques
Reference
in 4th International Conference on Data Management Technologies and Applications, 20-22 July, 2015, Colmar, Alsace, France (DATA 2015), ISBN 978-989-758-103-8, pp. 91-96, 2015
Description
Model tree is a useful and convenient method for predictive analytics in data streams, combining the interpretability of decision trees with the efficiency of multiple linear regressions. However, missing values within the data streams is a crucial issue in many real world applications. Often, this issue is solved by pre-processing techniques applied prior to the training phase of the model. In this article we propose a new method that proceeds by estimating and adjusting missing values before the model tree creation. A prototype has been developed and experimental results on several benchmarks show that the method improves the accuracy of the resulting model tree.
