Page 16 - Machine Learning for Subsurface Characterization

Chapter 1

Unsupervised outlier detection

techniques for well
logs and geophysical data

Siddharth Misra*, Oghenekaro Osogba †,a and Mark Powers ‡
*
Harold Vance Department of Petroleum Engineering, Texas A&M University, College Station,
† ‡
TX, United States, Texas A&M University, College Station, TX, United States, The University of
Oklahoma, Norman, OK, United States
Chapter outline
1 Introduction 2 4.4 Metrics/scores for the
1.1 Basic terminologies in assessment of the
machine learning and performances of
data-driven models 3 unsupervised ODTs on the
1.2 Types of machine learning conventional logs 21
techniques 3 5 Performance of unsupervised
1.3 Types of outliers 4 ODTs on the four validation
2 Outlier detection techniques 5 datasets 26
3 Unsupervised outlier detection 5.1 Performance on Dataset #1
techniques 7 containing noisy
3.1 Isolation forest 8 measurements 26
3.2 One-class SVM 8 5.2 Performance on Dataset #2
3.3 DBSCAN 10 containing measurements
3.4 Local outlier factor 11 affected by bad holes 28
3.5 Influence of hyperparameters 5.3 Performance on Dataset #3
on the unsupervised ODTs 12 containing shaly layers
4 Comparative study of and bad holes with noisy
unsupervised outlier measurements 29
detection methods on well logs 14 5.4 Performance on Dataset #4
4.1 Description of the dataset containing manually
used for the comparative labeled outliers 30
study of unsupervised ODTs 15 6 Conclusions 32
4.2 Data preprocessing 15 Appendix A Popular methods for
4.3 Validation dataset 17 outlier detection 33

a
Formerly at the University of Oklahoma, Norman, OK, United States
Machine Learning for Subsurface Characterization. https://doi.org/10.1016/B978-0-12-817736-5.00001-6
© 2020 Elsevier Inc. All rights reserved. 1