M. Tech. Information Networks / This dissertation reports on the development of an algorithm based on e-commerce matchmaking to select students. The student selection is an important strategic decision making task for the management of higher educational institutions and human resource departments of corporate organizations. The selection task has proven to be tedious mainly because of the decision process, the fundamental assumptions made and the level of accuracy achieved. The decision error that results from inaccurate student selection process is one source of skill mismatch. This work therefore seeks to improve the student selection accuracy using matchmaking approach. In doing so, e-commerce matchmaking method is applied to student selection and offered as an effective way of integrating cognitive and noncognitive skills to improve the selection accuracy. The efficacy of the matchmaking algorithm is demonstrated through a prototype implementation of the algorithm and specifically applied to university student selection.

In 1956, E. Michael proved his famous convex-valued selection theorems for l.s.c. mappings de ned on spaces with higher separation axioms (paracompact, collectionwise normal, normal and countably paracompact, normal, and perfectly normal), [39]. In 1959, he generalized the convex-valued selection theorem for mappings de ned on paracompact spaces by replacing \convexity" with \ -paraconvexity", for some xed constant 0 < 1 (see, [42]). In 1993, P.V. Semenov generalized this result by replacing with some continuous function f : (0;1) ! [0; 1) (functional paraconvexity) satisfying a certain property called (PS), [63]. In this thesis, we demonstrate that the classical Michael selection theorem for l.s.c. mappings with a collectionwise normal domain can be reduced only to compact-valued mappings modulo Dowker's extension theorem for such spaces. The idea used to achieve this reduction is also applied to get a simple direct proof of that selection theorem of Michael's. Some other possible applications are demonstrated as well. We also demonstrate that the -paraconvex-valued and the functionally-paraconvex valued selection theorems remain true for C 0 (Y )-valued mappings de ned on -collectionwise normal spaces, where is an in nite cardinal number. Finally, we prove that these theorems remain true for C (Y )-valued mappings de ned on -PF-normal spaces; and we provide a general approach to such selection theorems. / Thesis (Ph.D.)-University of KwaZulu-Natal, Westville, 2012.