∑jJ1∑pj1vm.PDF

Manuscript Number : PMET −34 Title: Simultaneous Two-Way Fuzzy Clustering of Multiple Correspondence Analysis This paper concerns a method that combines Multiple Correspondence Analysis (MCA) with fuzzy k-means in a unified framework in order to classify the object scores of respondents and the weights for variable categories simultaneously; and this under the restriction that a common cluster is to be identified for a set of relatively homogenous respondents and variable categories. The paper seems very well motivated. The results of this type of work are quite encouraging; in particular as demonstrated by two applications (“The TV Program Preference Data” and “The French Worker Survey Data”). The joint maps offered by the proposed method, seem much easier to interpret due to simultaneous treatment of cluster structures of both respondents and variable categories. Specific comments and suggestions for improvement follow. Firstly, as far as I can tell, the approach of the paper deals with a problem which is equivalent to minimizing the following optimization criterion: J K N   ∑ SSF −Z W ∑ ∑ umSSf −b  j 1 j j k 1 i 1 ki i k K J Pj vm SSw −b  ∗∗∗ ∑ ∑ ∑p 1 kjp jp k k 1 j 1 with respect to F, W T , b , u , and v , subject to F F  I, and j k ki kjp K K T T ∑k 1 uki  ∑k 1 vkjp  1,where SSM traceM M traceMM . We suppose that these constraints are used as “hints” to guide the algorithm to a useful solution. The constraints do not seem clear because M