Controlling feature selection in random forests of decision trees using a genetic algorithm: classification of class...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Feature selection is critical in classification tasks with high-dimensional data.
• Random forests are powerful ensemble methods for classification.
• Genetic algorithms offer robust optimization capabilities.

Purpose of the Study:

• To introduce GenForest, a method integrating genetic algorithms with random forests for enhanced feature selection.
• To evaluate GenForest's performance in a competitive, real-world classification challenge.

Main Methods:

• Developed GenForest, a procedure controlling feature selection within random forests using a genetic algorithm.
• Applied GenForest to classification problems in the Comparative Evaluation of Prediction Algorithms 2006 (CoEPrA) competition.
• Utilized a dataset with 5787 features per peptide for classifying type I MHC nonapeptides.

Main Results:

• GenForest achieved competitive rankings in the CoEPrA competition: 10/23 (Problem 1), 5/16 (Problem 3), and 9/16 (Problem 4).
• The method effectively handled high-dimensional data, classifying type I MHC nonapeptides.

Conclusions:

• GenForest provides an effective approach for feature selection in complex classification problems.
• The method's performance in the CoEPrA competition validates its utility for biological data analysis.