Onome1991/Facies_prediction

Employ SVM, KNN and RF classifiers for the prediction of lithofacies type in Wells in the OBOM field Niger Delta # Facies_prediction Employ SVM, RF, KNN classifiers for the prediction of lithofacies type in Wells in the OBOM field Niger Delta The oil and gas industry has its foundation in reservoir characterization and it expends massive manpower and time in lithofacies prediction which is its bedrock. Hence, this study involves the training of three machine learning algorithms namely SVM, RF and KNN with core QCed lithofacies interpretations to predict facies and test the most accurate classifier. The data used for the analysis consists of the 14 wells that penetrated the Obom field. The field consists of various lithologies interpreted from the core ranging from channel sands (CH), Lower Shoreface (LSF), Upper Shoreface (USF), Heterolith (HT) and Shales (SH). The analysis showed that the distribution of lithologies was 52.74% for SH, 22.69 for USF, 20.70 for CH, 2.75 for LSF and 1.11 HT. The dataset was then The overall net pay of the field was 47.3% which points to a very high potential for hydrocarbon accumulation. After deploying the quantile transform with Gaussian distribution for standardisation the three classifiers were trained and used for a blind test on well OBOM-4. It was observed that the RF classifier was the most effective in facies prediction with the highest overall accuracy of 87% and best in predicting continuous units as it achieved the highest F1 scores for most dominant classes of facies shales (0.93), channels (0.85) and upper shoreface (0.82), while SVM had the lowest facies prediction accuracy on the blind test with 0.40, and adjusted /adjacent facies prediction of 0.66 and lastly, the KNN achieved a blind test accuracy of 0.58 and a adjusted facies accuracy of 0.82. This results showed that the RF classifiers best predicts the intercalating lithology of the Niger Delta.