Comparison among dimensionality reduction techniques based on Random Projection for cancer classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• High-dimensional data, particularly in genomics, requires efficient dimensionality reduction techniques.
• Random Projection (RP) offers fast feature reduction but often suffers from lower classification accuracy.
• The increasing volume of big genomics data necessitates improved dimensionality reduction methods.

Purpose of the Study:

• To enhance the classification accuracy of Random Projection (RP).
• To investigate the effectiveness of combining RP with other dimensionality reduction methods like Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Feature Selection (FS).

Main Methods:

• Evaluated the performance of combined dimensionality reduction techniques on microarray and simulation datasets.
• Compared classification accuracy and running time of various combinations, including FS followed by RP and LDA followed by RP.
• Utilized datasets such as BC-TCGA for experimental validation.

Main Results:

• Feature Selection (FS) followed by Random Projection (RP) achieved a 14.77% improvement in classification accuracy on the BC-TCGA dataset compared to RP alone.
• Linear Discriminant Analysis (LDA) followed by RP increased classification accuracy by 13.65% on the same dataset.
• FS followed by RP demonstrated superior classification accuracy across most tested datasets.

Conclusions:

• Combining Feature Selection (FS) with Random Projection (RP) is a highly effective strategy for improving classification accuracy in high-dimensional data.
• The integration of LDA with RP also yields a more discriminative subspace, enhancing performance.
• These hybrid approaches offer a promising solution for real-time analysis of big genomics data.