Gene expression classification is crucial but challenging due to the high dimensionality of genomic data and overfitting risks. To address this, we propose BOLIMES, a novel feature selection algorithm that refines the feature subset for improved classification. Unlike traditional methods, BOLIMES combines the robustness of Boruta with the interpretability of LIME, ensuring only the most relevant genes are retained. It first uses Boruta to filter out non-informative genes and then applies LIME to rank the remaining genes based on their local importance. An iterative evaluation selects the optimal subset to maximize predictive accuracy. By combining feature selection with interpretability, BOLIMES effectively reduces dimensionality while maintaining high classification performance, offering a powerful solution for gene expression analysis.