<p>Lung diseases are among the many deadly illnesses and widespread worldwide. This paper proposes an optimized AlexNet architecture and improved Quasi-random Fractal Search based on Opposition-based Learning to classify numerous lung diseases. Opposition-based learning increases population diversity and prevents the population from sinking into a local optimum. Moreover, the proposed model’s capabilities extend to solving global optimization problems, demonstrated through testing on the IEEE Congress on Evolutionary Computation 2022 test suite. The results showed that the proposed model outperformed the original algorithm and the other eight metaheuristic algorithms according to statistical convergence, Friedman, and Bonferroni–Dunn tests. To demonstrate the overall effectiveness of the proposed model, it is then used to optimize the Hyperparameters of the AlexNet model (IQRFS-AlexNet) for lung disease classification. The obtained results showed that the IQRFS-AlexNet model outperforms the compared metaheuristic-based AlexNet model, achieving an overall accuracy of 99.01%, a sensitivity of 99.10%, a precision of 99.12%, a specificity of 99.24%, an F-score of 99.11%, and geometric Mean of 99.12% on six lung diseases x-ray datasets. The IQRFS-AlexNet model performed better than four other pre-trained transfer learning models.</p>