An efficient approach for gene selection through parallel bio-inspired algorithms and Shapley value analysis

The fast development of microarray technology has significantly assisted in the use of gene expression analysis to forecast cancer subtypes. Analyzing high-dimensional microarray data is still challenging, as existing hybrid methods cannot find highly discriminative genes. This study aims to use Shapley value analysis and hybrid bio-inspired algorithms to develop a scalable, parallel gene selection technique to increase computing efficiency and classification accuracy in high-dimensional microarray data. This study used hybrid feature selection approaches inspired by bio-organisms to create a scalable parallel gene selection system. The dataset size is initially enlarged by Adaptive Synthetic Sampling (ADASYN). They use the Recursive Feature Elimination (RFE) approach to extract features and determine their Shapley values. In addition, the Whale Optimization Algorithm (WOA) works to determine which genes are most important. After that, Machine Learning (ML) techniques assist in classifying the chosen characteristics. According to the experiment results, the suggested strategy surpasses standard gene selection techniques with the same datasets, employing improved classification accuracy and reducing computing time. K-NN achieved an accuracy of ۸۵.۴۴%, while LR showed improved results with an accuracy of ۹۱.۷۲%. RF further increased accuracy to ۹۴.۶۹%. SVM demonstrated exceptional performance, reaching an accuracy of ۹۷.۶۳%. Ultimately, XGBoost excelled among all models with the highest accuracy of ۹۸.۴۹%, highlighting its robust ability to classify SRBCT samples effectively based on gene expression data.

کلیدواژه ها:

Microarray technology ، High-Dimensional Data ، Machine Learning ، Shapley Value Analysis ، Bio-inspired algorithms