A collection of resources and commentary providing an introduction to supply chain management and related systems for students, practitioners, and anyone else interested in learning more about how to design, manufacture, transport, store, deliver, and manage products.
Wednesday, September 11, 2013
Data Mining in SCM: Use of Bayesian Networks in Inventory Management
Combining Prior knowledge with data analysis
Effective inventory management is a crucial part of the
supply chain of many organizations. Big Corporations such as Wal-Mart are
always praised for maintaining a low inventory turnover while not facing shortage
To maintain their inventory at an acceptable level, SCM
executives use either forecast demand based on historical data or scheduled
requirements demand based on known customers’ orders. There are many other
techniques to manage inventory: renew inventory when if falls below certain
threshold, holding a safety stock in case the demand cannot be precisely
One would argue that all these techniques can give
desirable results but present some limitations because new demand may be
different from historical demand due to external factors and clients can cancel
orders due to change in the economy or a shift in their own customers’ demand.
When those changes can be predicted either by data, expertise or inquiry, a
company would probably benefit from combining their data with such analysis so
that they do not face disastrous situations.
A Bayesian network is the use of prior knowledge along
with data to determine the outcome of the combination of certain factors. Even
though partially based on beliefs or expertise knowledge, such networks have
proven to be very reliable .
Two years ago, I was working for a major shipping
line company and one of the challenges we were facing was the effective
handling of our containers and stuffing products’ inventory. The inventory
management was crucial because we did not want to have idle containers that
would incur daily storage charges while they could be used in some places such
as China where the demand was high. At the same time, we had 80% of the local market
and would not want our customers to the competition just because we did not
have enough containers or stuffing material.
This was challenging because we had customers that
shipped seasonally (cashew nuts exporters or toys importers for Christmas and New
Year) but their volume was closely dependent on external factors. For instance,
the cashew nuts volume will be impacted by the local price that can be determined
by the rainy season or the inflation and the global market demand that can be changed
by a bad season in Brazil for example or a low demand in India. Thus when
forecasting the volume, we would not only rely on the forecast demand based on
historical volume and clients’ previsions but also on our knowledge of all external
factors that can affect the overall volume.
Partial representation of a BN, without probabilities
We were informally using a Bayesian network to
determine the probability of the volume being low or high based on our prior
knowledge of the probabilities and conditional probabilities associated to
every combination of events. For example, a political instability would lead to
inflation on prices and we would know that the exporters could not afford to
ship their forecasted volume.
that many companies use belief networks in their supply chain management but
they fail to do it professionally by implementing an information system that
would incorporate it into their SCM system. Such system would incorporate all the
variables and can even get better by learning form constant data addition.
The Bayesian network is just an example of a data
mining technique that can be added to traditional inventory management systems.
With the development of analytics and BI, the companies are given new
innovative ways to enhance their supply chain management and this might be the
key differentiator between a good SCM and a great one.