A non-parametric learning algorithm for small manufacturing data sets

doi:10.1016/j.eswa.2006.09.008

2008-01-01

PRE

Der‐Chiang Li *

Chun-Wu Yeh

DOI：10.1016/j.eswa.2006.09.008

原文链接

原文求助

摘要

En 中文

Nowadays the manufacturing environment changes promptly owing to globalization and innovation. It is noteworthy that the life cycle of products consequently becomes shorter and shorter. Although data mining techniques are widely employed by researchers to extract proper management information from the data, scarce data can only be obtained in the early stages of a manufacturing system. From the view of machine learning, the size of training data significantly influences the learning accuracies. Learning based on limited experience will be a tough task. On account of the cause, this research systematically estimates the data behavior such as the trend and potency to capture the dependency within a sequence of time series data. It should also be added that the analyzed data in this article are dependent examples that come from different populations. This research proposes a non-parametric learning algorithm instead of using parametric statistics for small-data-set learning. The proposed algorithm named the trend and potency tracking method (TPTM) attempts to explore the predictive information through the generation of trend and potency (TP) value of each datum. The extra information extracted from the data trend and potency proves that it can speed up stabilizing the learning task and can dynamically improve the derived knowledge from the occurrence of the latest data. (c) 2006 Elsevier Ltd. All rights reserved.

Keyword:

trend and potency function

small data sets

non-parametric learning

machine learning