Title
Allison Ronis
Kaplan University IT521-2: Analyze business intelligence systems
Background
Coors Brewers Ltd., in the United Kingdom is proud to be one of the top beer brands. There are many variables a person has when choosing a beer, taste, mood, venue and occasion. Coors wants to make sure that when a customer chooses which brand to drink they are choosing Coors. The company believes that being creative will bring long-term success and having the ability to determine the changing moods of people and sell beer accordingly will bring in profits. (Turban, Sharda, & Delen, 2011, p. 282, para. 2)
One important issue with beer is flavor. Typically, the flavor is determined by test panels. These tests are usually time-consuming. Coors wants to understand the chemical composition of flavors and if they knew that, it would open doors that have not been opened yet. “The relationship between chemical analysis and beer flavor is not clearly understood yet” (Turban, Sharda, & Delen, 2011, p. 282, para. 3). There is data on sensory analysis and chemical composition and Coors needs a way to link them together. The answer was Neural networks.
Neural Networks
The simplist defination of Neural Networks, more referred to as an ‘Artificial neural network’ (ANN) is defined by Dr. Robert Hecht-Nielsen, as a “computer system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs” (A Basic Introduction To Neural Networks, n.d.). With different interconnected layers such as many input layers, one output layers; one operation called the training operation, one could add many inputs (variables) into the system to reach the desired outcome. However at first it must be said that in order to reach the desired outcome, many inputs have to be entered over and over again and changing the neorons each time, in order to reach the desired outcome. (Stergiou & Siganos, n.d. 3.1, A simple neuron). So how could this help Coors?
Coors has collected much data over the years and used this data with the Neural networks to gain insite on how to get creative with their beers and ensure people choose their beers over other beers on the market. They took some data and first created a single neural network to see the relationship between sensory and analytical data flavor and one single quality. NeuroDimension, Inc. (nd.com) provided the package solution for Coors. “The neural network consisted of an MLP architecture with two hidden layers” (Turban, Sharda, & Delen, 2011, p. 282, para. 4). Typical architecture of an MLP with two hidden layers. (Geronimo, Carlos, Fernando, Paulo, & Eduardo, 2013).
By using many inputs and outputs the neural network trained and learned the relationship between these inputs and output combinations. The training was terminated automatically when it had been observed there were no improvements in network errors in the last 100 epochs. During the training sessions, there were many randomizations of inputs to ensure there would not be any unfairness.
This method did not work due to two factors. The first was that because they concentrated on a single quality meant the difference in the data was low. It was impossible for the neural network to determine the relationships from the data. Secondly, they found that “only one subset of the provided inputs would have an impact on the selected beer flavor” (Turban, Sharda, & Delen, 2011, p. 283, para. 6). The determination was that there was too much noise within the inputs that it had no impact on the flavor. They needed a better way.
Coors deciced to use a genetic algorithm. This algorthm was able to control different switches for the input variables and minimize the errors in the neural network. “When this minimum was reached, the switch settings would identify the analytical inputs that were most likely to predict the flavor” (Turban, Sharda, & Delen, 2011, p. 283, para. 8). The results were, they were able to determine which inputs were useful and identify which flavors could be predicted better. In the end the neural network was able to predict flavors by using chemical inputs. It is hard to determine sensory responses with algorithms or systems. There are too many variables.
Some benefits to this kind of information going forward is the determination of what people like and what they would buy. Neural networks are only going to get better from here. If successful Coors could supply consumers with beers that were flavored according to their mood, occasion, and venue.
I would consider, although time consuming, taste tests and good old fashioned polling to see what is liked or not. I believe in order to get the best results you need to combine human interactions with computer programs and computer programs are not the end all be all. Works Cited
A Basic Introduction To Neural Networks. (n.d.). Retrieved August 18, 2014, from Cs.Wisc.edu: http://pages.cs.wisc.edu/~bolo/shipyard/neural/local.html
Geronimo, T. M., Carlos, E. C., Fernando, d. C., Paulo, R. A., & Eduardo, C. B. (2013, January 16). MLP and ANFIS Applied to the Prediction of Hole Diameters in the Drilling Process. Retrieved August 18, 2014, from Intech: http://www.intechopen.com/books/artificial-neural-networks-architectures-and-applications/mlp-and-anfis-applied-to-the-prediction-of-hole-diameters-in-the-drilling-process
Stergiou, C., & Siganos, D. (n.d.). NEURAL NETWORKS . Retrieved August 18, 2014, from http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html
Turban, E., Sharda, R., & Delen, D. (2011). Decision Support and Business Intelligence Systems (9th ed.). Upper Saddle River, New Jersey: Prentice Hall.