Feature Selection Boosted by Unselected Features


The goal of feature selection is to identify those characteristics that are most strongly relevant and to eliminate the others. In recent times, there has been a lot of interest in embedded feature selection methods. These methods incorporate learning of feature weights into the process of training a classifier, which has garnered a lot of attention. Traditional embedded methods, on the other hand, only concentrate on the combinatorial optimality of all of the features that are chosen. They will often choose the features that are only marginally relevant because their combination abilities are satisfactory, but they will omit some features that are extremely relevant, which will bring the generalization performance down. In order to solve this problem, we have developed a brand new embedded framework for feature selection that we call feature selection boosted by unselected features (FSBUF). To be more specific, we modify the conventional embedded model by adding an additional classifier for features that have not been chosen, and we jointly learn the feature weights in order to maximize the classification loss caused by features that have not been chosen. As a consequence of this, the additional classifier reuses the strongly relevant features that were not selected to replace the features in the selected feature subset that were only moderately relevant. Our ultimate goal can be expressed as an optimization problem called a minimax optimization problem, and we devise an efficient gradient-based algorithm to solve it. In addition, we provide a theoretical demonstration that the proposed FSBUF is capable of enhancing the generalization ability of conventional embedded feature selection methods. Extensive testing on both artificial and actual-world data sets demonstrates that FSBUF provides superior performance while also being easily understandable.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here

PROJECT TITLE :Visualization, Selection, and Analysis of Traffic FlowsABSTRACT:Visualization of the trajectories of moving objects results in dense and cluttered images, that hinders exploration and understanding. It also hinders
PROJECT TITLE : Distributed Mobile Sink Routing for Wireless Sensor Networks A Survey - 2014 ABSTRACT: The concentration of data traffic towards the sink in a wireless sensor network causes the nearby nodes to deplete their
PROJECT TITLE :A Fast Clustering-Based Feature Subset Selection Algorithm for High-Dimensional Data - 2013ABSTRACT:Feature selection involves identifying a subset of the most useful features that produces compatible results as

Ready to Complete Your Academic MTech Project Work In Affordable Price ?

Project Enquiry