American Oil and Gas Reporter - April 2016 - 78

SpecialReport: Drilling Technology
it can be useful to compute the ranks
using different predictive models to get a
more robust sense of which predictors
are important. In the case of the Wolfcamp
dataset, there is a great deal of disagreement among the models as to which predictors are most influential. Depth is popular among all models. Three of the four
models also put weight on amount of
proppant, lateral length, and the amount
of fracture fluid.

Variable importance rankings can differ
widely by method. It can be useful to try
several methods to gain more robust feedback about which variables are most influential in driving the response. However,
different methods have different limitations
and considerations.
For example, using Akaike or Bayesian
criterion as measures of variable importance can be problematic in situations
where predictors are highly correlated. If

INSTRUCTOR - Led

COURSES

t IADC WellSharp™
Drilling/Workover

tIADC WellCAP®

Workover/Well Servicing

tIWCF Rotary Drilling
- Learning

COURSES

tIADC WellSharp™
Drilling Operations

tIADC WellCAP®

Workover/Well Servicing

tWCS Certified Drilling
tWCS Basic Rig Math
Register today:
www.wellcontrol.com

+1.713.849.7400
www.wellcontrol.com

78 THE AMERICAN OIL & GAS REPORTER

a pair of predictors happen to be correlated,
one predictor can compensate when the
other is removed, resulting in only a
small reduction in the model fit. As a
result, both correlated predictors appear
unimportant.
Indeed, this is the case in the Wolfcamp
dataset for some predictor pairs. For example, the amounts of frac fluid and
proppant pumped are highly correlated
(as can be observed in the pair-wise scatter-plot in Figure 1). This underlines the
importance of performing a preliminary
exploratory data analysis before jumping
into model building.
Random forest and gradient boosting
models have custom methods for identifying influential predictors. In RF, each
variable's prediction strength is measured
by calculating the increase of MSE when
a given variable is permuted while all
others are left unchanged. The rationale
is that if a predictor is not influential, rearranging its value among the training
observations will not significantly change
the model's prediction accuracy. For GBM
models, variable importance is based on
the number of times a predictor is selected
for splitting, weighted by the squared
improvement to the model as a result of
each split, and averaged over all trees.
There is a growing trend toward using
statistical and machine learning techniques
in oil and gas applications. The goal is to
mine big data and develop data-driven insights to understand and optimize reservoir
performance. But statistical analysis is
still at an early stage of maturity. Petroleum
engineers and geoscientists need to develop
a better understanding of the full repertoire
of available techniques and their potential.
At the same time, data scientists need to
understand the types of decision-making
problems specific to unconventional reservoirs, and to apply appropriate statistical
learning techniques for developing robust
data-driven insights.
❒

Coming In May
The Conventional Reservoir Renaissance series looks at independents' "off-the-radar" vertical drilling
projects in conventional geologic formations in the Mid-Continent region.
AOGR also makes the rounds at
leading operators in the SCOOP,
STACK and CNOW unconventional
trends in Oklahoma to update the latest field developments.

American Oil and Gas Reporter - April 2016

Table of Contents for the Digital Edition of American Oil and Gas Reporter - April 2016