Efficient Techniques Based on Sparse Representation for Classifying High-dimensional Multiclass Microarray Data

Abstract

Sparse representation (SR) has shown strong performance in classification tasks, particularly for high-dimensional data such as microarray gene expression profiles. These datasets present significant challenges due to their high dimensionality and limited sample size, which often hinder the performance of conventional classifiers. Methods: SR addresses this by expressing each signal as a linear combination of a small subset of training samples, reducing computational complexity and improving accuracy. However, using all training samples in the dictionary increases computational cost. This study explores several SR-based classifiers to address microarray data classification, focusing on dictionary construction strategies and sparse coding algorithms. Results: Experimental results on the 14-Tumors dataset show that selecting a subset of representative atoms and applying the SL0 algorithm significantly improves both speed and classification accuracy. Conclusions: These findings highlight the potential of SR approaches for effective and efficient classification of high-dimensional biological data.