Categorising Insurance Policy Data with MLPs and SOMs

A. Shah and C. Huyck (UK)


SOM, Categorisation, Data mining, MLP, Knowledge Discovery, Prediction, Customer Relation management


Two connectionist techniques are applied to real world insurance policy data. Multi Layer Perceptrons are trained to categorise data using a variant of back propagation called Back Percolation. Self Organising Maps are also used; as they are a clustering algorithm, a semi-supervised algorithm is used to modify the system so that it categorises. On this particular data set, the modified SOM system gets 196 out of the 238 positive results significantly outperforming the Multi-Layer Perceptron, which gets 35. Moreover, the SOM system outperforms the best prior system, a Bayesian model, which got only 121.

Important Links:

Go Back