A Modified Cultural Algorithm for Feature Selection of Biomedical Data

Abstract

An important step in developing predictive models is determining the best features to be used for building the models. Feature selection algorithms are frequently adopted to remove non-informative and redundant features from the dataset before building the predictive models. However, it can be a significant challenge to determine the features which would make the best predictors, particularly in large datasets. Importantly, building a model with the best subset of features can make it more interpretable and efficient. This paper proposes a novel feature selection algorithm which is based on the cultural meta-heuristic optimization algorithm. The modified Cultural Algorithm was developed and optimized for achieving high accuracy and Area Under the Curve. The quality of the selected features was assessed using the performance of a Support Vector Machine classifier. The proposed algorithm was tested on five benchmark datasets and achieved an average accuracy of 0.923 and an Area Under the Curve of 0.898 across all datasets. Although the performance of the proposed modified Cultural Algorithm was comparable with that of the standard Genetic Algorithm, the modified Cultural Algorithm required a smaller number of features and shorter execution time than the Genetic Algorithm.