An efficient approach for feature construction of high-dimensional microarray data by random projections.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Dimensionality reduction of high-dimensional microarray data presents significant computational challenges.
• Genetic programming (GP), while powerful, struggles with performance on high-dimensional datasets.
• Exploring GP's generalization capabilities requires effective dimensionality reduction techniques.

Purpose of the Study:

• To investigate the efficacy of random projections (RPs) for dimensionality reduction in microarray data.
• To evaluate the performance of GP when combined with RP-constructed features.
• To compare the performance of RP-GP with other machine learning algorithms using both full and reduced features.

Main Methods:

• Applied random projections (RPs) to reduce the dimensionality of eight diverse microarray datasets.
• Integrated RP-constructed features with a genetic programming (GP) approach.
• Benchmarked the RP-GP method against decision trees, random forest, naive Bayes, support vector machines, and k-nearest neighbors using identical feature sets.

Main Results:

• Features constructed using RPs demonstrated superior performance when utilized with the GP approach.
• The combined RP-GP method showed significant improvements in model generalization and computational efficiency.
• Seven of the eight datasets used were novel to machine learning research, providing new benchmarks.

Conclusions:

• Random projections offer a computationally efficient method for dimensionality reduction in high-dimensional data.
• Combining random projections with genetic programming enhances model performance and generalization capabilities.
• The proposed approach provides a robust alternative for analyzing complex biological datasets.