The diagnosis of lung cancer in primary stages is essential for providing appropriate treatment and enhancing survival rates. This paper presents a novel approach to identifying lung cancer into two subtypes, adenocarcinoma and mesothelioma, using microarray gene expressions. The research employs the fast Walsh–Hadamard transform (FWHT) for dimensionality reduction and utilizes two bio-inspired feature selection methods, harmonic search (HS), and the dragonfly algorithm (DA). The investigation involves five machine learning classifiers, namely Gaussian mixture model (GMM), Naïve Bayes classifier (NBC), random forest (RF), decision tree (DT), and support vector machine (SVM) with a radial basis function (RBF). The random forest classifier reported the highest accuracy of 90.05% when features were selected using the dragonfly algorithm, alongside a good detection rate of 89.15%. This demonstrates the method's potential for early lung cancer detection and highlights the importance of gene expression data in distinguishing between subtypes.