Pyramidal Gravitational Search clustering for Hybrid Gene Expression Data

Since many real-world performance concerns are becoming bigger, more complex, and more dynamic, natural-inspired algorithms are becoming more and more popular. Due to the scale and complexity of the problems, it is necessary to develop an optimization strategy whose effectiveness is measured by finding the nearly optimal solution in a reasonable amount of time. The bio-inspired metaheuristic techniques are based on swarm intelligence, biological, physical, and chemical features, depending on the source of inspiration. Cancerous tumors form and propagate largely as a result of genetic factors. A high-dimensional data processing difficulty exists with the identification of the pertinent genes. The Pyramid-GSA hybrid framework was developed to tackle the issue of dimensionality reduction through a cyclical gene-reduction method. This framework integrates two advanced modules: a high-throughput filtering component and an iterative wrapper technique inspired by the gravitational search algorithm (GSA). Using a recursive selection approach, top-performing genes are retained in each reduction cycle, facilitating efficient dimensionality compression while preserving essential genetic features. This methodology aims to optimize the gene subset to enhance classification accuracy and minimize interference from non-informative genes. Testing was conducted on a multi-class breast cancer microarray gene expression dataset, with comparisons to established feature selection methods to ensure a comprehensive analysis. The various approaches are assessed and contrasted using datasets from the UCI repository using a set of assessment metrics. Findings demonstrate that in spite of initialization and the use of stochastic operators, for the most favorable feature combination, the suggested version of GSA aims to search the feature space for the optimal feature combination.