The geochemistry of basaltic rocks is widely used to investigate the tectonic setting of magmatism. The limitation of traditional two-dimensional tectonic discrimination diagrams is mainly risen from the fact that they can only simultaneously use the information of two (x-y plots) or three (ternary diagrams) elements (or element ratios) for discrimination. This obstacle can be overcome with the assistance of machine learning method, which shows great performances in classification of multidimensional datasets. In this study, we present a neural network-based model that uses whole rock major and trace elements to discriminate basaltic rocks (SiO₂ 45–55 wt%) from a wide range of tectonic settings, including continental arc basalt (CAB), island arc basalt (IAB), intra-oceanic arc basalt (IOAB), mid-ocean ridge basalt (MORB), oceanic plateau basalt (OPB), oceanic island basalt (OIB), continental flood basalt (CFB) and continental rift basalt (CRB). Using a modified method of cross validation, it is estimated that the model can discriminate the tectonic setting with an average accuracy of ∼86%, and ∼98% in discriminating the major tectonic regimes (arc, spreading center, or within-plate magmatism). This discrimination model was programed as a stand-alone Microsoft Excel spreadsheet that can be directly used by pasting the whole rock data into it. The discriminator was then applied to investigate the geodynamic background of the Paleoproterozoic (∼1.75 Ga) Xiong'er volcanism in the southern margin of the North China Craton (NCC). It has long been debated whether this magmatism took place in a continental arc or within-plate rift environment. The discrimination result shows that both the Xiong'er Group volcanic rocks and coeval intrusive rocks have CFB affinities, indicating that they were products of mantle plume activity and defines a large igneous province (LIP) in a within-plate setting. This, along with previous studies, constrains the breakup of NCC, a part of the Columbia supercontinent, not later than ∼1.79 Ga, and supports the idea that the fragmentation of the Columbia was triggered by mantle plume impingement.

玄武岩的地球化学被广泛用于研究岩浆作用的构造环境。传统二维构造判别图的局限性主要是因为它们只能同时使用两个（xy图）或三个（三元图）元素（或元素比率）的信息进行判别。借助机器学习方法可以克服这一障碍，该方法在多维数据集的分类中显示出了出色的性能。在这项研究中，我们提出了一种基于神经网络的模型，该模型使用整个岩石主成分和微量元素来区分玄武岩（SiO ₂45-55 wt％）的构造环境，包括大陆弧玄武岩（CAB），岛弧玄武岩（IAB），大洋内弧玄武岩（IOAB），中洋脊玄武岩（MORB），海洋高原玄武岩（OPB），大洋洲玄武岩（OIB），大陆洪水玄武岩（CFB）和大陆裂谷玄武岩（CRB）。使用改进的交叉验证方法，估计该模型可以判别构造背景，平均准确度约为86％，而辨别主要构造类型（弧，展布中心或板内岩浆作用）的准确度约为98％。 。该识别模型被编程为独立的Microsoft Excel电子表格，可以通过将整个岩石数据粘贴到其中来直接使用。然后将鉴别器应用于研究古元古代（〜1.75 Ga）的雄性地球动力学背景。华北克拉通（NCC）南部边缘的火山喷发。长期以来一直争论这种岩浆作用是在大陆弧还是板内裂谷环境中发生的。判别结果表明，熊耳群火山岩和近代侵入岩均具有CFB亲和力，表明它们是地幔柱活动的产物，并在板内设置了较大的火成岩省（LIP）。这与以前的研究一起，在不晚于〜1.79 Ga的情况下，限制了NCC（哥伦比亚超大陆的一部分）的破裂，并支持了哥伦比亚的分裂是由地幔柱撞击引起的。判别结果表明，熊耳群火山岩和同代侵入岩均具有CFB亲和力，表明它们是地幔柱活动的产物，并在板块内界定了一个大的火成岩省（LIP）。这与以前的研究一起，在不晚于〜1.79 Ga的情况下，限制了NCC（哥伦比亚超大陆的一部分）的破裂，并支持了哥伦比亚的分裂是由地幔柱撞击引起的。判别结果表明，熊耳群火山岩和同代侵入岩均具有CFB亲和力，表明它们是地幔柱活动的产物，并在板块内界定了一个大的火成岩省（LIP）。这与以前的研究一起，在不晚于〜1.79 Ga的情况下，限制了NCC（哥伦比亚超大陆的一部分）的破裂，并支持了哥伦比亚的分裂是由地幔柱撞击引起的。