Journal of Applied Sciences1812-56541812-5662Asian Network for Scientific Information10.3923/jas.2017.204.211MustaphaMuhammad Firdaus B. KhalidNoor Elaiza Bt Abd IsmailAzlan B. 42017174Background: Self-organizing map (SOM) is a very popular algorithm that has been used as clustering algorithm and data exploration. The SOM consists of complex calculations where the calculation of complexity depending on the circumstances. Many researchers successfully improve SOM processing speed using discrete Graphic Processing Units (GPU) since the introduction of Compute Unified Device Architecture (CUDA) in 2007 and Open Computing Language (OpenCL) in 2009. In spite of excellent performance using GPU, there are performance issues in processing a large mapping size especially dealing with find the Best Matching Unit (BMU) and updating weightage. Additionally, the larger mapping size also could burden the processing through the usage of high memory capacity which leads to high rate memory transfer. Recently, heterogeneous systems, that soldered CPU and GPU together on a single chip are rapidly attractive the design paradigm for todays platform because of their remarkable parallel processing abilities. Therefore, this study evaluates parallel SOM performance on discrete GPU and heterogeneous system in order to improve the algorithm processing. Materials and Methods: Accordingly, this study demonstrates parallel SOM that comprises of three kernels. The parallel SOM then executes on two different platforms: (1) Discrete GPU platform and (2) Heterogeneous system platform. This study evaluates the outcomes of the computation experiments based on computation time and SOM quality measurements. Results: As a result, parallel SOM that executed on heterogeneous system platform is able to reduce the total processing time compared to discrete GPU platform when processing large mapping sizes and large data sets. Conclusion: More important, this study highlights how the proposed parallel SOM can improve the execution performance and maintain the SOM results when running on heterogeneous system.]]>Kohonen, T.,2013375265Arribas-Bel, D., K. Kourtit and P. Nijkamp,201331248257Perelygin, K., S. Lam and X. Wu,201440241251Wittek, P. and S. Daranyi,201373198206Kirk, D.B. and W.W. Hwu,20132nd Edn.,Pages: 514Pages: 514Gaster, B.R. L. Howes, D. Kaeli, P. Mistry and D. Schaa,20122nd Edn.,Mukherjee, S., Y. Sun, P. Blinzer, A.K. Ziabari and D. Kaeli,20162016Khronos OpenCL Working Group,20142014Hasan, S., S.M. Shamsuddin and N. Lopes,20146114De, A., Y. Zhang and C. Guo,2016198180189Richardson, T. and E. Winer,20158817Lachmair, J., E. Merenyi, M. Porrmann and U. Ruckert,2013112189199Mustapha, M.F.B., N.E.B. Khalid and A.B. Ismail,20152015pp: 8492Khan, S.Q. and M.A. Ismail,20132013pp: 233237Catanzaro, B.,2010TM optimization case study: Simple reductions.]]>2010Davidson, G.,20152015Kyriazis, G.,20122012Chattopadhyay, M., P.K. Dan and S. Mazumdar,201212600610