Study of the Application of Disagreement-based Collaborative Regression in Log Interpretation

Abstract

In the field of log interpretation, it’s easy to acquire a lot of data, however, it requires cost to get the label information, thus the labeled samples are often not enough. The secondary interpretation of porosity is a typical application task with few labeled samples and a large number of unlabeled samples. Manual interpretation has the disadvantages of strong subjectivity and low accuracy. A disagreement-based co-training style semi-supervised regression algorithm was proposed as an alternative to the manual interpretation. Two kNN regressors with disagreement were employed. Each of them labels unlabeled samples with high confidence level for the other to improve regression estimates. The method was verified through the experiments which showed that the generalization performance of this semi-supervised model is better than the other supervised models in such cases.