Genetic Programming for Feature Selection and Classification

Genetic programming (GP) is an evolutionary computation paradigm that evolves computer programmes or symbolic expressions through operations analogous to natural selection and genetic variation. In the context of feature selection and classification, GP serves two complementary roles. First, as a feature selector it explores subsets of original variables by treating each candidate subset as an individual and assessing its predictive fitness against a base classifier. This wrapper-style approach balances predictive accuracy and model parsimony via multi-objective fitness functions that penalise both misclassification and excessive feature count. Second, as a feature constructor it generates novel variables by combining or transforming original features using arithmetic, logical or nonlinear operators, thereby uncovering complex relationships that may elude traditional methods. When integrated with classifiers such as support vector machines, decision trees or neural networks, GP-derived feature pipelines have demonstrated improved discrimination in domains ranging from genomics to cybersecurity and affective computing. Key strengths of GP include its ability to handle high-dimensional data, adaptively tailor representations to problem structure and yield interpretable, tree-based models amenable to human inspection.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent advances illustrate GP’s versatility in automating feature engineering. In cybersecurity, a multi-tree GP framework was devised to construct composite features that adapt to evolving threat patterns, achieving notable gains in balanced accuracy for advanced persistent threat detection. In data mining and machine learning, a GP-based feature ranking method leveraged the frequency of feature appearances within evolved trees combined with a multi-criteria fitness function to produce compact, high-performance subsets across diverse benchmark datasets. In affective computing, GP-driven feature selection on multi-channel EEG data reduced an initial set of 70 features to a targeted subset of 32, enhancing emotion classification accuracy and offering interpretable insights into neural correlates of emotional states. These contributions underscore GP’s capacity to streamline feature pipelines and bolster classification performance in varied practical settings.

Genetic Programming for Feature Selection and Classification publication trend

The graph below shows the total number of articles in genetic programming for feature selection and classification across all publications each year (not limited to Nature Index journals).

Technical terms

Genetic Programming: An evolutionary algorithm that evolves programmes or expressions by simulating selection, crossover and mutation.
Feature Selection: The identification of a relevant subset of original variables to improve model generalisation and reduce dimensionality.
Feature Construction: The creation of new variables by transforming or combining existing features to capture non-linear patterns.
Wrapper Method: A feature selection approach that evaluates subsets based on the performance of a specific classifier.
Fitness Function: A quantitative measure combining classification accuracy and model complexity used to guide evolutionary search.
Multi-class Classification: Classification tasks involving more than two target classes, often addressed via one-against-one or one-against-rest strategies.

References

1. Genetic programming for enhanced detection of Advanced Persistent Threats through feature construction. Computers & Security (2025).
2. A feature selection method with feature ranking using genetic programming. Connection Science (2022).
3. Genetic Programming‐Based Feature Selection for Emotion Classification Using EEG Signal. Journal of Healthcare Engineering (2022).