Mukund Deshpande and Geroge Karypis. Using conjunction of attribute values for classification. In Proceedings of the 11th International Conference on Information and Knowledge Management, 2002.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....sets that only had two classes. Due to the performance on these data sets, I did not run the SVM classifiers on the multi class problems. Further, to verify that the performances of the SVMs are comparable to what others have reported, I compared to two other works using SVM on the UCI data [MM01, DK02] The performance of my SVM runs performed comparably to these, thus SVMs should not necessarily be the learner of choice on the data used in this study. 126 C4.5 Ripper Entropy wins 32 29 Numerical wins 49 41 Insignificants 69 80 Table 8.3: How often each type of classifier and encoding was ....

Mukund Deshpande and Geroge Karypis. Using conjunction of attribute values for classification. In Proceedings of the 11th International Conference on Information and Knowledge Management, 2002.

....large number of times. Once the complete set of such sub structures has been identified, our algorithm then proceeds to build a classification model based on them. To a large extent, this approach is similar in spirit to the recently developed frequent itemset based classification algorithms [29, 28, 8] that have been shown to outperform traditional classifiers that rely on heuristic search methods to discover the classification rules. The overall outline of our classification methodology is shown in Figure 2. It consists of three distinct steps: i) feature generation, ii) feature selection, ....

....first four problems (P1, P2, P3, and P4) derived from the PTC dataset, the performance actually improves with feature selection. Such improvements are possible even in the context of SVM based classifiers as models learned on lower dimensional spaces will tend to have better generalization ability [8]. Also note that for some datasets the number of features decreases as # increases. This is because the features that were selected have higher average support. Topological versus Geometric Subgraphs The various results shown in Tables 2 4 also provide an indication on the relative performance ....

Mukund Deshpande and George Karypis. Using conjunction of attribute values for classification. In Proceedings of the eleventh CIKM, pages 356--364. ACM Press, 2002.

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