Page 205 - Data Science Algorithms in a Week

Predictive Analytics using Genetic Programming 189

CONCLUSION

Our experience working with complex problems, incomplete data, and high noise
levels have provided us with a more comprehensive methodology where machine
learning base-models can be used with other types of empirical and exact models. Data
science is very popular in the marketing domain where first-principle models are not
common. However, the next frontier of big data analytics is to use information fusion -
also known as multi-source data fusion (Sala-Diakanda, Sepulveda & Rabelo, 2010). Hall
and Llinas (1997) define data fusion as “a formal framework in which are expressed
means and tools for the alliance of data originating from different sources, with the aim
of obtaining information of greater quality”. Information fusion is going to be very
important to create predictive models for complex problems. AI paradigms such as GP,
are a philosophy of the “data fits the model.” This viewpoint has many advantages for
automatic programming and the future of predictive analytics.
As future research, we propose combining GP concepts with operations research and
operations management techniques, to develop methodologies where the data helps the
model creation to support prescriptive analytics (Bertsimas & Kallus, 2014). As we see in
this paper these methodologies are applicable to decision problems. In addition, it is a
current tendency in the prescriptive analytics community to find and use better metrics to
measure the efficiency of the models besides the confusion matrix or decile tables.
Another important point for engineered systems is the utilization of model-based system
engineering. SysML can be combined with ontologies in order to develop better GP
models (Rabelo & Clark, 2015). One point is clear: GP has the potential to be superior to
regression/classification trees due to the fact that GP has more operators which include
the ones from regression/classification trees.

ACKNOWLEDGMENTS

We would like to give thanks to Dr. Bruce Ratner. Bruce provided the GenIQ Model
for this project (www.GenIQModel.com). In addition, we would like to give thanks to the
NASA Kennedy Space Center (KSC). KSC is the best place to learn about complexity.
The views expressed in this paper are solely those of the authors and do not
necessarily reflect the views of NASA.

REFERENCES

Bertsimas, D., & Kallus, N. (2014). From predictive to prescriptive analytics. arXiv
preprint arXiv:1402.5481.