Case Study on the Effectiveness Feature Engineering Techniques for Performance Improvement of Data Mining Applications

In this paper we try to demonstrate the effectiveness of different techniques for feature selection and optimization strategies that are used for data mining applications. Feature extraction is an important step in every data mining application, which aims to find discriminative patterns, interconnection, and structures within large datasets. In this paper we try to illustrate the effective use of the classical methods such as Principal Component Analysis (PCA), Kernel principal component analysis (KPCA), as well as more innovatory methods to better represent the data so as to achieve a better performance of the underlying pattern analysis, datamining tasks. We also consider effective representation of the data items in a discriminative way. We demonstrate this by making use of synthetic data so that a visual analysis can be made. We also use 3 real world datasets to demonstrate the effectiveness of PCA and KPCA based representations for obtaining improved performance for an SVM based multi-class classifier.