Nature-inspired computing based Non-Hodgkin lymphoma prediction from microarray expression with GA-RFE gene selection method; an experimental histopathological study

Non-Hodgkin lymphoma (NHL) is a cluster of cancer types in the blood, usually starting in the white blood cells called lymphocytes. It is a part of the immune system of the body. It affects the lymph system, otherwise the lymphatic system of the body. This system defends the body against infections and supports the movement of fluids throughout the body. In general, cancer development has a strong association with genetic factors. Molecular profiling would be more beneficial to find the root cause of disease progression. Technological advancements in healthcare have recently brought up effective disease diagnosis and treatment strategies. Automated disease diagnosis systems are profound and assist medical practitioners in better decision-making in critical situations. They have shown their prominence in precise disease diagnosis and possibly better medications for specific illnesses. Intelligent computational algorithms are the backbone for the models to find insights from the data. These patterns are the key factors in the group between different categories. But, finding the pattern from high- dimensional data is a highly challenging task. In this experimental histopathological study, an effective hybrid feature selection technique is proposed to address this issue method which fuses the nature- inspired genetic algorithm (GA), wrapper-based feature subset selection and recursive feature elimination (RFE) technique. The proposed GA-RFE model finds 35 biomarkers as an optimal feature subset inputted into supervised machine learning classification algorithms for training and evaluation. The performance of the models on the subset is validated through evaluation metrics. The proposed system attained better accuracy, 96.8% higher than the benchmarked algorithms and proved its efficacy in biomarker selection. These biomarkers are trained with supervised classifiers to find the best model under different performance evaluation metrics.