A probabilistic neural network (PNN) based framework for lithology classification using seismic attributes.

This paper proposes a Probabilistic Neural Network (PNN) based framework for classification of lithology from a number of seismic attributes. The PNN has been the natural choice for classification in several research areas due to its insensitivities towards outliers and higher computational speed compared to multilayer perceptron (MLP) networks. Initially, the lithology is labelled into four classes such as sand, shaly sand, sandy shale, and shale through thorough analysis of multiple well logs by a proficient geologist. The seismic attributes and well logs pertaining to twelve closely spaced boreholes from a western onshore hydrocarbon field in India are used in this study. The performance of the designed framework consisting of pre-processing, classification, and lithological maps generation stages is compared with existing supervised classifiers in terms of classification accuracy, sensitivity, and specificity and the results are reported. The selection of appropriate parameters associated with individual classifier and importance of individual seismic predictors are also investigated. Finally, lithology maps indicating the different classes are produced using the tuned parameters of PNN over the study area. This framework would be of immense help to geologists along with other geological measures to estimate the probability of the presence of hydrocarbon in a large study area. • A framework having pre-processing, classification and lithological maps generation. • Generation of lithology maps over the study area. • Comparison with supervised classifiers using accuracy, sensitivity, and specificity. • Included results of hyperparameter tuning for different classifiers. • Investigation on the relative importance of the predictor variables. [ABSTRACT FROM AUTHOR]