This paper presents an empirical analysis of using deep learning features with an Extreme Learning Machine (ELM) for automatic modulation classification in wireless communication. The study utilizes advanced pre-deep learning models like VGG 16, Resnet 50, and Inception V3 to extract features, which are then fed to an ELM for classification. The ELM’s performance is enhanced by the moth flame optimization method (MFOP-ELM). To measure the efficacy of the proposed model, it is tested on two standard datasets - RADIOML2016.10A and RADIOML 2018.01A. The paper also compares the proposed model with other classifiers such as SVM, K-NN, and BPNN. Results show that the proposed model using Inception V3 features with the MFOP-ELM classifier outperforms other state-of-the-art models.