DenseNet architectures have demonstrated impressive performance in image\nclassification tasks, but limited research has been conducted on using\ncharacter-level DenseNet (char-DenseNet) architectures for text classification\ntasks. It is not clear what DenseNet architectures are optimal for text\nclassification tasks. The iterative task of designing, training and testing of\nchar-DenseNets is an NP-Hard problem that requires expert domain knowledge.\nEvolutionary deep learning (EDL) has been used to automatically design CNN\narchitectures for the image classification domain, thereby mitigating the need\nfor expert domain knowledge. This study demonstrates the first work on using\nEDL to evolve char-DenseNet architectures for text classification tasks. A\nnovel genetic programming-based algorithm (GP-Dense) coupled with an\nindirect-encoding scheme, facilitates the evolution of performant char DenseNet\narchitectures. The algorithm is evaluated on two popular text datasets, and the\nbest-evolved models are benchmarked against four current state-of-the-art\ncharacter-level CNN and DenseNet models. Results indicate that the algorithm\nevolves performant models for both datasets that outperform two of the\nstate-of-the-art models in terms of model accuracy and three of the\nstate-of-the-art models in terms of parameter size.\n