DATA MINING WITH NEURAL NETWORKS
by Rich Edwards, Marketing Director, Trajecta, Inc.
Many different technologies and techniques are employed in today's data mining world. One of the most advanced methodologies utilized is artificial neural networks.
Artificial neural networks are capable of modeling extremely large and complex problems which may involve hundreds of variables. The models they generate can identify patterns and relationships in data that were previously unknown. Neural networks are non-parametric and are capable of taking on any form that the data requires. This ability makes neural networks extremely accurate predictors of non-linear real world events.
Neural networks were largely inspired by the way the human brain processes information. Neural networks utilize databases of historical information to make predictions and inferences to be applied to future marketing efforts. These efforts could be directed at the same database of customers (a new offer) or can be generalized to a new database of potential customers.
Neural networks start with an input layer of nodes, each of which represents an independent variable such as customer demographics, psychographics, or transaction records. These inputs are connected to a layer of nodes in a hidden layer. There may be numerous hidden layers. Finally, outcomes such as predicted ROI, response to a marketing campaign, profitability, etc. are produced in an output layer.
As inputs pass through the network to the output layer, predicted results are compared to actual results and the error is measured. Next, through a process called backpropagation the network sends the results back through the network and adjustments are made. Each pass through of the network is called an epoch. This process continues until the error is minimized to a point in which a generalization to a larger population can be accurately made. Thus, similar to the way humans process information, as the neural network "trains" itself on the data, it "learns" the relationships that exist between the independent and dependent variables.
---
For more information, see http://www.trajecta.com