Kohonen Network Modeling for Asteroid Name Recognition

Abstract

A small number of asteroids have already had their own permanent names. However, tracking the names of asteroids that have been used before is an impractical work because there are thousands of names that must be traced one by one. This research intends to minimize the search burden of many asteroid name data using the Kohonen network. By using the Kohonen network, it is sufficient to do training on the sample data provided which is far less than the actual data. The result of this training is then used to obtain the number of asteroid names that are successfully identified by the Kohonen network. The result can also be used to propose a new asteroid name so that thestatus of acceptance of the proposed name can be determined. Based on the results of the training on the sample data, the training result is getting better as the learning rate increases. However, when tested with real data, the overall result that is not satisfactory is obtained because the level of recognition is only 49.78%. From the test result, it is also found that there is no linear relationship between the level of learning rate and the number of names that were successfully identified. Further research that can be done are the inclusion of non-asteroid training data, changing Kohonen network parameters, or using other recognition methods.