Assessing performances of unsupervised and supervised neural networks in breast cancer detection

This paper deals with the comparison of the two neural network methods of learning: supervised (classical feedforward neural networks: multi-layer neural networks (MLP), radial basis function (RBF) and probabilistic neural networks (PNN)) and unsupervised (self organizing feature maps (SOFM), or Kohonen map), in order to assess their performances on a labeled breast cancer database. By revealing their equivalence on such a complete database (i.e. including both input and output), it is to be expected that in a real-world situation of a non-labeled database (i.e. patients without previous diagnosis), only the unsupervised method represented by SOFM will be able to make a good decision without the benefit of a supporting teacher.