Graph-based neural networks, including transformers, are the most used neural networks. Recently, overlapping community detection problems have been tackled using graph convolutional networks (GCN), achieving superior performance than classical methods. However, some of these methods have limited practical use as they require ground truth information to optimize the hyper-parameters. In addition, most existing works, including GCN-based methods, either ignore the edge weight of weighted graphs or fail to leverage the edge weight information to achieve better performance. To this end, we propose PRODG as a practical overlapping community detection method in edge-weighted graphs using deep GCNs. Specifically, we design weighted dilated aggregation algorithms to incorporate the information of edge weights to achieve a deep-weighted residual GCN (WResGCN) model. Moreover, we perform a completely unsupervised modularity-based depth search for deep GCNs, making our method available for practical applications. We apply our model to real-world Facebook graphs with synthetic edge weights, having reliable ground truth information. Experimentation and evaluation show that our methods effectively leverage the edge-weight in weighted graphs. Our completely unsupervised depth search strategy in deep GCNs performs competitively or better than state-of-the-art methods, bolstering the practical use of deep GCNs in overlapping community detection.