Search

Storage device performance prediction is a key element of self-managed storage systems and application planning tasks, such as data assignment and configuration. Based on bagging ensemble, we proposed an algorithm named selective bagging classification and regression tree (SBCART) to model storage device performance. In addition, we consider the caching effect as a feature in workload characterization. Experiments indicate that caching effect added in feature vector can substantially improve prediction accuracy and SBCART is more precise and more stable compared to CART. [ABSTRACT FROM AUTHOR]

Context prediction, highlighted by accurate location prediction, has been at the heart of ubiquitous decision support systems. To improve the prediction accuracy of such systems, various methods have been proposed and tested; these include Bayesian networks, decision classifiers, and SVMs. Still, greater accuracy may be achieved when individual classifiers are integrated into an ensemble system. Meanwhile, General Bayesian Network (GBN) classifier possesses a great potential as an accurate decision support engine for context prediction. To leverage the power of both the GBN and the ensemble system, we propose a GBN-assisted ensemble system for location prediction. The proposed ensemble system uses variables extracted from Markov blanket of the GBN΄s class node to integrate GBN, decision tree, and SVM. The proposed system was applied to a real-world location prediction dataset, and promising results were obtained. Practical implications are discussed. [ABSTRACT FROM AUTHOR]

The aggregation of multiple unstable classifiers often leads us to reduce the misclassification rates substantially in many applications and benchmark classification problems. We propose here a new variant of the double bagging, where we use the support vector machine as the additional classifier built on the out-of-bag samples. The underlying basic classifier is the decision tree. We use four kernel types; linear, polynomial, radial basis and sigmoid kernels, expecting the new classifier perform better. The major advantages of the proposed method is that, 1) it has robustness against many messy real data cases, 2) the generation of support vectors in the first phase facilitate the decision tree to classify the objects with higher accuracy, resulting a significant reduction in misclassification rates in the second phase. We also used subsamples with bootstrap samples, where 50% samples are used for the training samples without replacement, expecting larger out-of-bag samples. We have applied the proposed method to a real case, the condition diagnosis for the electric power apparatus; the feature variables are the maximum likelihood parameters in the generalized normal distribution, and these variables are composed from the partial discharge patterns of electromagnetic signals by the apparatus. Comparing to other well-known ensemble classifiers, the double bagging with the support vector machine classifier with radial basis kernel performs best among all the classifiers. [ABSTRACT FROM AUTHOR]

Partially synthetic data comprise the units originally surveyed with some collected values, such as sensitive values at high risk of disclosure or values of key identifiers, replaced with multiple draws from statistical models. Because the original records remain on the file, intruders may be able to link those records to external databases, even though values are synthesized. We illustrate how statistical agencies can evaluate the risks of identification disclosures before releasing such data. We compute risk measures when intruders know who is in the sample and when the intruders do not know who is in the sample. We use classification and regression trees to synthesize data from the U.S. Current Population Survey. [ABSTRACT FROM AUTHOR]