M. Dash, H. Liu, J. Yao. Dimensionality Reduction of Unsupervised Data. Proceedings of the 9th International Conference on Tools with Arti cial Intelligence (ICTAI'97).  Home/Search   Document Details and Download   Summary   Related Articles   Check

....amount of research in this area, the vast majority of the research has been carried out under the supervised learning paradigm (supervised feature selection) paying little attention to unsupervised learning (unsupervised feature selection) Only a few works exist addressing the latter problem. In [6], the authors present a method to rank features according to an unsupervised entropy measure. Their algorithm works as a lter approach plus a backward sequential selection search. Devaney and Ram [8] proposes a wrapper approach combined with either a forward or a backward sequential selection ....

....standard accepted performance task for supervised learning to unsupervised learning. Whereas the former learning comprises the prediction of only one feature, the class, from the knowledge of many, the latter aims the prediction of many features from the knowledge of many [12] On the other hand, [6, 8, 40] evaluate their unsupervised fea9 ture selection mechanisms by measuring the class label predictive accuracy of the learnt models over the cases of a testing set after having performed learning in a training set where the class labels were masked out. The speed of learning and the ....

[Article contains additional citation context not shown here]

M. Dash, H. Liu and J. Yao, \Dimensionality Reduction for Unsupervised Data," Proceedings of the Ninth IEEE International Conference on Tools with AI, IEEE Computer Society Press, pp. 532-539, 1997.

....a more general and principled mechanism that inspired this work. Fisher et al. [5] adapted Gennari s procedure to a diagnosis task, where the intent was to minimize the number of probes necessary to diagnose a fault. As in supervised learning, preprocessing approaches are more common as in [3], 4] or [13] However, neither of these works have been extensively evaluated along all the dimensions proposed here. As regard the exible prediction task, the only existing work is [16] with a weak evaluation and our own work in preprocessing and postprocessing methods [14, 15] Although the ....

M. Dash, H. Liu, and J. Yao. Dimensionality reduction for unsupervised data. In Ninth IEEE International Conference on Tools with AI, ICTAI'97, 1997.

....experimental evidence, their method appears to work fairly well. However, since they used the metric of a particular clustering system to evaluate the importance of features, it remains unclear if this method can be extended to work with other algorithms. Another promising proposal appears in [5], where an unsupervised entropy based measure for ranking features is described. However, the empirical evaluation is carried out by comparing the selected features with the features selected by a supervised method and by using a supervised system. In order to assess the real capabilities of the ....

M. Dash, H. Liu, and J. Yao. Dimensionality reduction for unsupervised data. In Ninth IEEE International Conference on Tools with AI, ICTAI'97, 1997.

....speed up the retrieval process (see Fig. 1) Although this approach lacks a strong theoretical justification, it is worth comparing to the more rigorous methods. We show some results of this comparison in section 4. 9 3. 3 Entropy based Feature Ranking This method proposed by Dash and Liu [5,6] seems best suited to the query selection task. It uses an entropy measure to rank the features in order of importance allowing us to access the most important feature directly. In addition, this entropy measure does not need class information to evaluate the features, unlike the information ....

Dash M., Liu H., Yao J. (1997) Dimensionality Reduction for Unsupervised Data, in proceedings of IEEE International Conference on Tools with AI (TAI-97), pp. 532-539.

.... reduction or feature selection has been recognized as a central problem in data analysis (Fayyad, Piatetsky Shapiro, and Smyth, 1996) The importance of this problem is reflected in the significant attention that this topic has recently received in the literature (Blum and Langley, 1997# Dash and Liu, 1997# Kohavi and John, 1997) However, the vast majority of the research in feature selection has been carried out under the supervised learning paradigm, paying little attention to unsupervised learning problems. By contrast with supervised learning approaches, in unsupervised learning there are no ....

....there is little experimental evidence, their method appears to work fairly well. However, since they used the metric of the COBWEB system to evaluate the importance of features, it remains unclear if this method can be extended to work with other algorithms. Another promising proposal appears in Dash, Liu, and Yao (1997), where an unsupervised entropybased measure for ranking features is described. However, the empirical evaluation is carried out by comparing the selected features with the features selected by a supervised method and by using a supervised system. In order to assess the real capabilities of the ....

[Article contains additional citation context not shown here]

Dash, M., Liu, H., and Yao, J. (1997). Dimensionality reduction for unsupervised data. In Ninth IEEE International ConferenceonTools with AI, ICTAI'97.

....embedded in CLASSIT, a descendant of COBWEB, and made some preliminary experiments. However, his research differs from this work in that we focus in a complex flexible prediction task and not in predicting a single class label. Two works that apply feature selection as a preprocessing step are [2] and [3] but again, evaluation is performed over class labels (but see [14] for some results in a multiple prediction task using a preprocessing step) In sum, our work is novel in two aspects. First, the retrospective pruning view makes feature selection a postprocessing step rather than a ....

M. Dash, H. Liu, and J. Yao. Dimensionality reduction for unsupervised data. In Ninth IEEE International Conference on Tools with AI, ICTAI'97, 1997.

....decrease the effectiveness of learning algorithms, especially if most of these features appear to be irrelevant with regard to the learning task. In fact, feature selection is a central problem in inductive learning as suggested by the growing amount of research in this area (Blum Langley, 1997; Dash Liu, 1997; Kohavi John, 1997) Clustering is a form of unsupervised learning used to discover useful patterns in data. Particularly, hierarchical clustering methods construct a tree structured clustering where sibling clusters partition the observations covered by their parent. Hierarchical structuring ....

....selection similarly to supervised learning approaches, most of the work concerning feature selection has been carried out under the supervised paradigm, paying little attention to unsupervised learning tasks. Only few proposals exist concerning this problem, and to our knowledge only two works (Dash, Liu, Yao, 1997; Devaney Ram, 1997) have explored preprocessing mechanisms of feature selection. In this paper we present a study of unsupervised methods of feature selection applied as a preprocessing step. We discuss the particular benefits that feature selection may provide in hierarchical clustering tasks ....

[Article contains additional citation context not shown here]

Dash, M., Liu, H., & Yao, J. (1997). Dimensionality reduction for unsupervised data. In Ninth IEEE International Conference on Tools with AI, ICTAI '97.

No context found.

M. Dash, H. Liu, J. Yao. Dimensionality Reduction of Unsupervised Data. Proceedings of the 9th International Conference on Tools with Arti cial Intelligence (ICTAI'97).

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC