Абстрактный

Protein Structural Class Prediction Using Feature Elicitation and Classification

Abinaya Suky .S and Selvakumar .S

In this paper a hybrid approach is proposed for the feature elicitation and classification of protein structures. Attribute extraction involves simplifying the amount of resources required to describe a large set of data accurately. Prediction of protein structural class is defined as follows: all alpha, all beta, alpha + beta and alpha / beta. Pattern recognition based approaches are used for many of the enhancements. Sequence based and physicochemical based attribute extraction is used in the existing. In sequence based attribute extraction, there are two methods and they are evolutionary based composition feature group and evolutionary based auto covariance feature group. In the physicochemical based attribute extraction also has two methods and they are overlapped segmented distribution approach and overlapped segmented auto correlation. The physicochemical based attribute extraction is based on the consensus features. Fast correlation based filter algorithm is proposed and is to find out the symmetrical uncertainty. In that case, the best features are selected. Various classification algorithms are proposed for classifying the protein structures. They are AdaBoost.Ml, Logit Boost, Support Vector Machine (SVM), Naïve Bayes, and Multi Layer Perceptron (MLP). Based on these algorithms, the majority votings are validated to predict the protein structures.