Home // UBICOMM 2010, The Fourth International Conference on Mobile Ubiquitous Computing, Systems, Services and Technologies // View article

Particularized Cost Model for Data Mining Algorithms

Authors:
Andrea Zanda

Keywords: ubiquitous, data mining, cost model, algorithm

Abstract:
Ubiquitous devices demand autonomous and adaptive data mining. Despite some advances, the problem of calculating the cost associated to the execution of data mining algorithms is still a challenge. Thus, in this paper we provide a method for predicting the cost in terms of efficacy and efficiency associated to a mining algorithm, the resulting cost model as shown in our previous work can be exploited by a mechanism for predicting the best configuration of a mining algorithm according to context and resources. Recent work presents how a cost model not associated to any dataset can provide reliable estimations on efficiency and efficacy, here we present how we can improve the accuracy of such estimations by particularizing cost model to a predefined dataset. We provide the guidelines of the method and then we present a particularized cost model for C4.5 algorithm associated to a specific dataset (Parkinson’s tele- monitoring). Experimental results show how the particularized cost model achieves significant better estimations than the general cost model.

Pages: 79 to 84

Copyright: Copyright (c) IARIA, 2010

Publication date: October 25, 2010

Published in: conference

ISSN: 2308-4278

ISBN: 978-1-61208-100-7

Location: Florence, Italy

Dates: from October 25, 2010 to October 30, 2010