EFFICIENT UNSUPERVISED AND SUPERVISED LEARNING ALGORITHM USING LOCAL LINEAR EMBEDDING SYSTEM
Keywords:
Unsupervised learning and supervised learning, feature selection, relevance, redundancy, high dimensionality, clustering, feature selection.Abstract
Many fields of science depend on exploratory data analysis and data redundancy. The demand to analyze large amounts of multivariate data raises the central problem of dimensionality reduction. The Spectral feature selection identifies relevant features by measuring their capacity of preserving sample similarity. It offers a potent framework for both supervised and unsupervised feature selection, it has been shown to be in force in many realworld applications. The redundant features can receive substantial adverse effects on learning performance, it is necessary to address this limitation for spectral feature selected. In this paper I aim to carry out the spectral feature selection algorithm to handle feature redundancy, taking a hybrid model. In this report to conduct theoretical analysis of the attributes of its optimal solutions, develop a correlation-based method for relevance and removing redundancy analysis, and carry on an empirical survey of its efficiency and effectiveness comparing with representative methods.
