Research On Imbalance Data Classification And Its Application On Lithology Recognition Of Ground-layer Rock Data

The classification of imbalance data is one of the ten major challenges in the field of machine learning,which has attracted many scholars in recent years.After further studying and summarizing the current methods that cope with imbalance data classification,we proposed the improved methods from the views of data pre-processing and ensemble learning in this paper.And we also studied the method of lithology recognition of ground-layer rock data,which is the significant application field of imbalance data classification.The specific work of this article is described as follows:First,we proposed a pre-processing method for imbalance data based on the best balance ratio,which is short for obtio.By obtaining the best balance ratio of each specific imbalance data,the obtio algorithm we proposed improved the classification results from the view of data pre-processing.The experimental results showed that the F-measure value was increased by0.2 compared with the method proposed by GU Qiong.Second,we proposed a method based on cluster and bagging for imbalance data imbalance,which is short for C&B.By employing the cluster and bagging and combining the ensemble learning rules,the C&B algorithm we proposed improved the classification results from the view of ensemble learning.The experimental results showed that the F-measure value was increased by 0.15 compared with the method proposed by Wei-Chao Lin.Third,because of the low recognition rate of ground-layer rock data using traditional methods,we proposed a method for lithology recognition of ground-layer rock data,which is short for LRGL.The LRGL we proposed took full account of the characteristics of each sample among the ground-layer rock data at the stage of data pre-processing.By over sampling the minority rock samples belong to specific types,we get balanced ground-layer rock data.Then,BP neural network was employed to recognize the lithology.The experimental results showed that the AUC value was increased by 0.2 compared with the method proposed by MOU Dan,and the performance of lithology recognition of LRGL is also stable.