The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9649

Title:Neural Networks for Motor Insurance Rating
Date: 1996
Abstract:In this project, an automated motor insurance rating system was simulated through an inductive learning technique, C4.5, and a neural network, back-propagation and tested using real data. Their results were comparatively analysed. Furthermore, extensive experiments were performed to find the optimal parameters of back-propagation, which critically govern the performance of back-propagation. The experimental results obtained from the two learning algorithms were unexpectedly not satisfactory as a classifier. The poor results of the two learning methods are analysed with respect to the problems of collected data and learning algorithms. Apart from this, the practical tips to build neural systems suggested in many reports were analyzed through experiments. In this project, the basic concerns to build an automated classification systems are examined carefully rather than investigating the advanced techniques. Automated classification systems can be built more easily in the future based on this analysis of the practical and basic concerns.

[Search These Pages] [DAI Home Page] [Comment]