Weight tuning and pattern classification by self organizing map using genetic algorithm

The paper deals with supervised learning. In many problems, the training data contains only the final judgment information in conjunction with the input data, but in some problems, more information needs to be extracted from the training data. A typical example is a medical diagnosis. The objective of the paper is to give the user internal information contained in the data by using only the binary class-information data. A self organizing map (SOM) is used as the main tool for this purpose. Our method is to tune the weight of the elements of the data so that the data of the same category tend to be mapped in the near points on the SOM, and the separation of different categories can be carried out successfully. A genetic algorithm (GA) is used for the tuning of the weight coefficients. After the learning, we can obtain the feature map, as well as the weight coefficients of the elements that indicate the importance for the categorization for the current data.