PM Network - January 2017

Getting It Done PROJECT MANAGEMENT IN ACTION
Hari Doraisamy, PMP, is team lead of mis-
sion critical support, SAP, Newtown Square,
Pennsylvania, USA.

The Power of AI Predictions
Artificial neural networks can boost the accuracy of project estimates.
By Hari Doraisamy, PMP

neural network.

I designed a neural network and trained it with a data set from 27 different projects. Parameters included project duration, complexity of the project, number of resources and projected cost.

The set also included the actual cost for each of
these 27 scenarios. I trained the network using the
historical data by iteratively adjusting the weights
until the calculated output matched the desired
output within toler-
ance for each training
record.

When presented with
new data, my neural
network was able to
estimate the project
costs based on the his-
torical information. I
tested it with simple
sample projects and
found it impressively
accurate. For more
complex projects, the
network also did an
excellent job of estimat-
ing costs, though it was
not as accurate as with simple projects.

Neural networks have been widely used in science and engineering. Project managers have been slower to adopt them, possibly due to a lack of awareness. But neural networks have great potential for our profession. And although my experiment focused on project costs, they also could be used to forecast other project data. PM

It is notoriously difficult for project man- agers to accurately estimate project costs. However, software that uses neural net- works can be a powerful solution to this problem.

Artificial neural networks, a type of artificial
intelligence (AI), mimic the way neurons in the
brain work. They “learn” patterns based on histori-
cal data and can then estimate values that depend
on several inputs.

Project managers know that the cost
of a project largely depends upon
three variables: the number of
resources, their cost and the
project’s length. However,
lesser variables, like scope
expansion and complexities,
could also impact the proj-
ect’s cost. That’s where
artificial neural networks
come in—we can use
them to identify potential
cost overruns based
on data from past
projects of similar
size and type.

This can allow a
project manager
to more accurately
plan project activities
and resources—pos-
sibly helping to prevent
cost overruns.
Project managers
who want to give it a try
could use commercial
software to create a cus-
tomized neural network. A good programmer
also can develop a computer program to design a