Thesis, Undergraduate (BS Applied Mathematics)- U.P. Mindanao

Yang's dissimilarity coefficient for mixed-type data was modified using two different aggregating equations of De Carvalho. L? Eixample normalized dissimilarity coefficient for continuous attributes was used instead of Yang's dissimilarity coefficient. This modified Yang's dissimilarity coefficients were then employed on constructing hierarchical trees using single linkage, complete linkage and UPGMA on auto, heart and credit data. Single linkage clustering algorithm was found to give higher misclassifications on auto data. This is due to the fact that single linkage has a tendency to cause chaining phenomenon. The efficiency of the two modified dissimilarity coefficients was then tested based on their accuracy, entropy and purity. The first dissimilarity coefficient was found to give better improvement on the accuracy, entropy and purity of Yang's dissimilarity coefficients.