Optimizing cancer diagnostics with modified moth flame optimization in high-dimensional microarray data

In recent years, high-throughput technologies have led to exponential increases in the volume and complexity of cancer microarray data, which present significant challenges for effective feature selection. Feature selection plays an important role in identifying similar genes and reducing the dimensionality of the data, thereby enhancing the performance of cancer classification models. This paper suggested a feature selection approach that relied on a modified version of the Moth Flame Optimization (MFO++) algorithm tailored specifically for high-dimensional cancer microarray data. The proposed MFO++ algorithm incorporates adaptive mechanisms influenced by the natural behavior of moths, such as attraction, repulsion, and random motion, to iteratively search for an optimal feature subset. Unlike traditional feature selection approaches, MFO++ dynamically adjusts the exploitation and exploration capabilities for the search process based on the characteristics of the dataset, thus improving the efficacy and efficiency of feature selection. Specifically, MFO++ achieved an average accuracy improvement of 6.5% across eight high-dimensional cancer gene expression datasets when compared with the standard Moth Flame Optimization (MFO) algorithm. This improvement was evaluated using classification accuracy as the primary metric, and the baseline for comparison was the performance of five classifiers (DT, RF, K-NN, SVM, and BPNN) without any feature selection or using standard MFO for feature selection.