همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

A New LDA-based Genetic Algorithm for Feature Selection and Classification in Gene Expression Data Analysis

نویسندگان :

Parisa Pouyamanesh¹ Mahdi Vasighi²

1- Institute for Advanced Studies in Basic Sciences (IASBS) 2- Institute for Advanced Studies in Basic Sciences (IASBS)

کلمات کلیدی :

Genetic Algorithm،Feature Selection،Linear Discriminant Analysis،Gene Expression Data

چکیده :

Recent advancements in cancer diagnosis have driven the need for more efficient genetic data analysis methods to improve the accuracy and speed of identifying influential genes. Genetic algorithms, inspired by evolutionary processes in nature, have emerged as powerful tools for optimizing and identifying cancer-related genes. However, a significant challenge in these algorithms is the gradual reduction of population diversity, leading to premature convergence on suboptimal solutions, a phenomenon known as genetic drift. This effect causes certain areas of the search space to be overlooked, ultimately reducing the accuracy of results. This study introduces a repair operation as part of the genetic algorithm’s process to address these limitations. This repair operation purposefully and intelligently eliminates irrelevant genes while enhancing essential ones, thus improving the quality of feature selection and preventing premature convergence. This approach uses Linear Discriminant Analysis (LDA) as a crucial tool to guide feature selection precisely. LDA coefficients, which reflect each feature's contribution to class discrimination, are used in a repairing mechanism applied to the individuals. This way, the proposed repair mechanism removes noisy and irrelevant features, preserving only those most influential in distinguishing between healthy and cancerous samples. This approach not only enhances diagnostic accuracy but also simplifies the model by finding more sparse candidate subsets, reducing the risk of overfitting. The proposed repair operation was applied to several gene expression benchmark datasets in terms of convergence speed and the performance of the proposed approach compared with the conventional GA-based feature selection. The results show that the proposed approach produces consistently better subsets with few gens and classification accuracies.