Flexible learning tree augmented naïve classifier and its application

Abstract

Tree augmented naïve Bayes classifier (TAN) has been widely used in machine learning and data mining. To improve the flexibility and classification performance of TAN, this paper proposes a Flexible Tree Augmented Naïve Bayes classifier (FTAN). In the FTAN, the mutual information contribution rate is used to measure the dependencies between attributes in the process of building the maximum weighted spanning tree. Then, a flexible filtering method is adopted to filter out edges with weak dependencies between attributes by dynamically adjusting the threshold. A range of experiments on UCI datasets reveals that the FTAN exhibits considerable advantages over other popular algorithms in terms of the 0-1 loss and class probability root mean square error. The FTAN is used to solve the problem of favourable distribution area prediction for the remaining oil and gas resources of the Jurassic Sangonghe Formation in the Junggar Basin. The application results show the effectiveness and superiority of the FTAN method, favourable areas of oil and gas resources are selected based on the FTAN prediction results. This provides a decision-making basis for optimising drilling strategies and oil and gas exploration targets.