SVM Tree for Personalized Transductive Learning in Bioinformatics Classification Problems
Personalized modelling joint with Transductive Learning (PTL) uses a particular local modelling (personalized) around a single point for classification of each test sample, thus it is basically neighbourhood dependent. Usually existing PTL methods define the neighbourhood using a (dis)similarity measure, in this paper we propose a new transductive SVM classification tree (tSVMT) based on PTL. The neighbourhood of a test sample is built over the classification knowledge modelled by regional SVMs, and a set of such SVMs adjacent to the test sample are aggregated systematically into a tSVMT. Compared to a normal SVM/SVMT approach, the proposed tSVMT, with the aggregation of SVMs, improves classifying power in terms of accuracy on bioinformatics database. Moreover, tSVMT seems to solve the over-fitting problem of all previous SVMTs as it aggregates neighbourhood knowledge, significantly reducing the size of the SVM tree.
KeywordsPersonalized Modelling SVM SVMT transductive learning transductive reasoning
Unable to display preview. Download preview PDF.
- 1.Vapnik, V.N.: The Nature of Statistical Learning Theory, 2nd edn., pp. 237–240, 263–265, 291–299. Springer, Berlin (1999)Google Scholar
- 2.Verma, A., Fiasché, M., Cuzzola, M., Iacopino, P., Morabito, F.C., Kasabov, N.: Ontology based personalized modeling for type 2 diabetes risk analysis: An integrated approach. In: Leung, C.S., Lee, M., Chan, J.H. (eds.) ICONIP 2009, Part II. LNCS, vol. 5864, pp. 360–366. Springer, Heidelberg (2009)CrossRefGoogle Scholar
- 5.Schölkopf, J.C., Platt, J.C., Shawe-Taylor, J., Smola, A.J., Williamson, R.C.: Estimating the support of a high-dimensional distribution. Technical report, Microsoft Research, MSR-TR-99-87 (1999)Google Scholar
- 7.Joachims, T.: Transductive Inference for Text Classification using Support Vector Machines. In: Procs of the Sixteenth International Conference on Machine Learning, pp. 200–209 (1999)Google Scholar