Optimization and data mining for fracture prediction in geosciences

The application of optimization and data mining in databases in geosciences is becoming promising, though still at an early stage. We present a case study of the application of data mining and optimization in the prediction of fractures using well-logging data. We compare various approaches, including multiple regression analysis (MRA), back-propagation neural network (BPNN), and support vector machine (SVM). The modelling problem in data mining is formulated as a minimization problem, showing that we can reduce an 8-D problem to a 4-D problem by dimension reduction. The MRA, BPNN and SVM methods are used as optimization techniques for knowledge discovery in data. The calculations for both the learning samples and prediction samples show that both BPNN and SVM can have zero residuals, which suggests that these combined data-mining techniques are practical and efficient.