Fused lasso for feature selection using structural information

Most state-of-the-art feature selection methods tend to overlook the structural relationship between a pair of samples associated with each dimension, which may encapsulate useful information for refining performance selection. Moreover, they usually consider candidate relevancy equivalent selected relevancy, and therefore, some less relevant features be misinterpreted as salient features. To overcome these issues, we propose new method based on graph-based representations Fused Lasso framework in this paper. Unlike approaches, our has two main advantages. First, it can accommodate through representation. Second, enhance trade-off individual one hand its redundancy pairwise other. This is achieved use applied reordered basis their relevance respect target feature. effectively solve optimization problem, an iterative algorithm developed identify most discriminative Experiments demonstrate that proposed approach outperform competitors benchmark datasets.