Using Evolutionary Algorithms For Hyperparameter Tuning And Network Reduction Techniques To Classify Core Porosity Classes Based On Petrographical Descriptions

Classifying the porosity of sedimentary information is an important field of study with applications to tasks such as oil reservoir characterisation. Classifying porosity into groups based on Petrographical characteristics has been attempted in the past using: expert systems, supervised clustering techniques and neural networks. In this paper, we expand upon the usage of neural networks for this classification task by applying Evolutionary Algorithms to determine optimal parameters. Despite recent advances in techniques to select hyperparameters it is still difficult to determine the optimal parameters for a given dataset. We further apply network reduction techniques to further improve classification accuracy. We produce results similar to the work done by Gedeon et al. [1] on this dataset.