A comparative study of the cost–beneft strategy with the learning ensembles of decision stumps in polymetallic prospectivity modelling
A comparative study of the cost–beneft strategy with the learning ensembles of decision stumps in polymetallic prospectivity modelling
复制标题
多金属前景建模中成本效益策略与决策树桩学习集成的比较研究
DOI：
10.1007/s12145-021-00709-z
复制
发表时间：
2021
影响因子：
2.8
通讯作者：
Tan Yulei
中科院分区：
文献类型：
--
作者：
Chen Yongliang;Zhang Yuanqing;Tan Yulei
找矿成本效益策略是一种易于实现的矿产远景建模算法，它采用似然比、提升指数和约登指数来表示矿产潜力。将找矿成本-效益策略扩展为利用马修斯相关系数（MCC）和F测度来表征矿产潜力，并与决策树丛组合Bagging和Boosting方法在青海省拉陵枣霍地区多金属找矿预测中进行了比较。在Bagging算法中用决策树替换决策树，可以缓解决策树深度引起的Bagging集成模型过拟合问题。根据扩展找矿成本效益策略的多金属找矿远景模拟结果，采用Bagging和Boosting集成方法，制作了7张成矿潜力图，包括似然比图、提升指数图、Youden指数图、MCC图、F测度图、分类得分图。受试者工作特征（ROC）曲线表明，在多金属找矿远景建模中，找矿成本效益策略优于集成学习模型的上级效果。根据7幅成矿远景图，优选出研究区多金属远景区。这些多金属远景区只占整个研究区的一小部分（12.61 - 16.95%），但几乎包含所有已知的多金属矿床（94 - 100%）。因此，在找矿成本效益战略中，可以用MCC和F测度来表征矿产潜力。扩展的找矿成本效益策略在多金属找矿远景建模中的应用效果优于集成学习算法。
The prospecting cost–benefit strategy is an easy-to-implement mineral prospectivity modeling algorithm which uses the likelihood ratio, lift index and Youden index to represent the mineral potential. In this study, the prospecting cost–benefit strategy was extended to use the Matthews correlation coefficient (MCC) and F-measure to represent the mineral potential, and compared with the bagging and boosting ensembles of decision stumps in polymetallic prospectivity modeling in the Lalingzaohuo district (Qinghai Province, China). Replacing the decision trees with decision stumps in the bagging algorithm can alleviate the overfitting problem of the bagging ensemble model caused by the depth of the decision trees. According to the polymetallic prospectivity modeling results of the extended prospecting cost–benefit strategy, and bagging and boosting ensembles, seven mineral potential maps were produced, including likelihood ratio map, lift index map, Youden index map, MCC map, F-measure map, classification score maps. The receiver operating characteristic (ROC) curves show that the prospecting cost–benefit strategy is superior to the ensemble learning models in polymetallic prospectivity modeling. According to the seven mineral potential maps, polymetallic prospective areas were optimally delineated in the study area. These polymetallic prospective areas account for only a small percentage of the whole study area (12.61 – 16.95%) but contain almost all known polymetallic deposits (94 – 100%). Therefore, the MCC and F-measure can be used to represent the mineral potential in the prospecting cost–benefit strategy. The extended prospecting cost–benefit strategy performs better than the ensemble learning algorithms in polymetallic prospectivity modeling.