Page 146 - Intelligent Digital Oil And Gas Fields

Components of Artificial Intelligence and Data Analytics 109

interpreted in the decision-making process. Analyzing large data sets is
inefficient and useless without powerful visualization tools and tech-
niques (explained further in Section 4.2.3). Hence, descriptive analytics
leverages heavily the concepts of exploratory data analysis (EDA)
(Gelman and Hill, 2007; Seltman, 2015), which integrates advanced
and interactive charting and graphing with concepts of univariate statis-
tics (e.g., plotting statistical distributions and modes, such as histograms,
distributions, mean, variance, confidence intervals, etc.), bivariate statis-
2
tics (e.g., cross plots, Q-Q plots, box plots, R , covariance statistics, and
spatial variogram-based analysis) and multivariate statistics or analysis
(MVA) (Tabachnick and Fidell, 2013), which combines analytical and
visualization techniques such as principal component analysis (PCA),
factor analysis, multidimensional scaling (MDS), or data clustering.
• Diagnostic analytics helps us determine the “root cause” of certain out-
comes. While traditional key performance indicators (KPI) can provide
a quantitative measure of performance, getting additional insight into
“why something happened” requires diagnostic business intelligence
tools. However, diagnostic analytics is laborious and frequently bur-
dened with hindsight bias (choosing data that matches results); it pro-
vides an improved understanding of a limited piece of the problem
we want to solve. For example, we can build an analytics dashboard
for providing information on the root-cause analysis of electric submers-
ible pump (ESP) failure events. The dashboard could show the basic lin-
ear causal relationships among variables; however, it would fall short in
capturing complex nonlinear variable correlations. Studies show that
<10% of companies surveyed do this type of analysis on occasion, and
<5% do so consistently.
The next two classes or types of data analytics are usually the ones referred to
by the technical and business analytics experts as the analytics types that can
really provide the insight and foresight into how to drive the technical and
business decisions forward. Predictive and prescriptive analytics are consid-
ered branches of the so-called cognitive analytics, because they combine ele-
ments, methods, and tools of cognitive science, such as AI, statistical
inference, ML, and multimodal deep learning (DL) of visual, language,
and knowledge recommendations.
• Predictive analytics provides the ability to use data (structured and unstruc-
tured) to derive patterns and forecast future events and outcomes
with mathematical certainty. It helps us discover hidden, nonintuitive
patterns in Big Data and understand complex causal relationships and