Page 189 - Data Science Algorithms in a Week

Predictive Analytics using Genetic Programming 173

COMPLEXITY AND PREDICTIVE ANALYTICS

Advanced analytics aims to provide the base necessary to handle complex problems
in terms of scalability and amount of data and sources (Chen & Zhang, 2014). The
analysis of the data is the new scientific paradigm, besides empirical, theoretic and
computational science. Challenges that create techniques and methodologies are
beneficial for this purpose in order to handle complex problems (Chen & Zhang, 2014).
A complex problem usually features at least several of the followings:

 Incomplete or lack of data,
 Very large amounts of data (i.e., petabytes)
 Hybrids of continuous and discrete variables/environments,
 Mix of structured and unstructured data,
 High noise levels
 Real-time, timeless, and latency features of the decision time window,
sensors/actuators system (to receive feedback and act), and the computational
execution of the predictive system
 Mix of qualitative and quantitative assessments
 Multidisciplinary and interdisciplinary features of the problem/system
 Nonlinearities, observability, and controllability issues
 Human-factors and human-behaviors (e.g., predictably irrational, usability,
culture, politics, etc.)

Our experience working and analyzing these problems have provided us with a more
comprehensive methodology where several models can be used with other types of
empirical models in order to build predictive systems. Our methodology has been
evolving through the years due to the technological trends mentioned above (i.e.,
computing power and new, more established AI techniques) and have the following steps
(Rabelo, Marin, & Huddleston, 2010):

1. Understand the problem from different viewpoints: We have to understand
the problem and the goals and objectives assigned to the predictive modeling
task. We have to view complex problems from different dimensions. This is
usually a multi-disciplinary/interdisciplinary effort. Some of the important
viewpoints are:
a. Basic Theory – First principles are very important to understand. The
team must be networked with the scientists and experts from the different
domains. The predictive modeling team has to be conversant with the
contributions of the different disciplines involved (materials, optics,
finance, marketing, human behavior, psychology, etc.).