miRNAs, as a vital component of the human genome, have a significant impact on various aspects of human health and well-being. Investigating unknown miRNA-disease associations can not only assist in treating certain diseases but also contribute to extending human lifespan and increasing happiness. However, traditional biological experiments to explore these associations consume substantial resources in terms of manpower and materials. Therefore, there is an urgent need to conduct extensive research on improved algorithms for effectively predicting unknown miRNA-disease associations. In this study, we introduce a novel approach named AEETCMDA for the prediction of miRNA-disease associations that are currently unknown. AEETCMDA leverages the HMDD v2.0 dataset as the primary data source and utilizes the Gaussian kernel function to calculate Gaussian interaction profile kernel similarities for both miRNAs and diseases. Additionally, stacked autoencoders are employed to extract latent features for miRNAs and diseases. Finally, the integrated data is fed into extremely randomized trees to predict unknown miRNA-disease associations. AEETCMDA is evaluated using five-fold and ten-fold cross-validation methods, achieving AUC values of 94.23% and 94.26% respectively. Furthermore, breast cancer and lung cancer are used for validation, and the top ten ranked miRNAs are validated with 9 and 10 miRNAs respectively.