Feature selection is an important process in various data relevant to the domains like financial organizations, marketing, healthcare, education etc. Identifying the relevant feature is the major impact in obtaining the accurate results. Optimal feature selection plays a vital role in predictive analysis. Feature selection or dimensionality reduction process are useful in identifying the most important variables or parameters which is useful in predicting the outcome. Feature selection helps to train the machine algorithm faster as well as reduced the complexity of the model. Classification plays an important role to identify the pattern analysis. Classification algorithms helps to group the major categories of data sets in supervised learning. Even though there are many number of attributes are available, only some features helps to improve the accuracy of the model. In the real world scenario filter methods are used to extract the features without outliers. Nevertheless Multicollinearity problem is not overcome by the filtering techniques. So that there is lack of identifying the best subset of features for better analysis. Existing systems use various learning algorithms (Feature Selection Algorithms) like fast forward selection and backward Elimination algorithm, Recursive feature elimination approach, Focus and relief algorithms. Principle component analysis (PCA) is well known technique to form the finest subset of features with the Neural Network classifier in supervised learning techniques. To improve the accuracy further we have proposed the new approach Borrowed PCA as an optimal feature selection method to interpret the best subset of features by using artificial neural networks classifier. And the method has been validated with the KDD Cup 1999 dataset used for measuring intrusion detection problems.
Volume 11 | 07-Special Issue
Pages: 846-857