A Hybrid Simple Filter-Wrapper Feature Selection Approach for Microarray Classification

High-dimensional datasets, characterized by a large number of features (p) and a limited number of samples (n), pose a significant challenge for machine learning tasks, particularly classification. This phenomenon, known as the “curse of dimensionality,” leads to several difficulties in achieving accurate classification. Microarray data, generated from gene expression profiling experiments, provides valuable insights into various biological processes. However, these datasets are inherently high-dimensional, making them susceptible to the curse of dimensionality. This study introduces a hybrid filter-wrapper attribute selection method to tackle the challenges of classification in high-dimensional datasets. The method employs the Gini impurity criterion to evaluate each attribute and reduce the dataset to varying percentages. Subsequently, a range of classification algorithms is applied to the reduced datasets. Additionally, a simple wrapper method is applied to the top 1% of attributes selected by the filter method. Evaluation of the proposed method reveals that selecting the top 1% and 5% of attributes yields the best classification results, considering balanced accuracy and 5-fold cross-validation. Further improvement in classification performance is achieved by applying the wrapper method. Notably, the proposed method exhibits minimal impact on execution time, a crucial aspect for practical applications.