| The self-organizing map (SOM) is a data mining technique that projects a high-dimensional input data space onto (usually) a two-dimensional grid visualization space. The mapping roughly preserves the most important topological and metric relationships among the original data and inherently clusters the data. In this thesis, the pattern recognition and selection of mutual funds are investigated by using existing, as well as improved, SOM technology. Results show that the SOM is an efficient tool for clustering and pattern recognition of mutual funds. The SOM Toolbox is improved by implementing an algorithm to provide explicit clustering that better matches the U-matrix visualizations than existing clustering algorithms. Additional post-clustering descriptive summaries and boxplots are provided to enhance interpretation of the SOM. The proposed improved SOM also provides a better basis than traditional methods for both mutual fund portfolio management and mutual fund performance comparisons. |
