Aging of a fuel cell (FC) is an unavoidable process, nevertheless managing operating conditions and performing timely maintenance or control can prolong its life span. More precisely, the prognostics of FC is major area of focus nowadays. This paper presents a data-driven approach for prognostics of Proton Exchange Membrane Fuel Cell (PEMFC) stack using constraint based Summation-Wavelet Extreme Learning Machine (SW-ELM). The proposition aims at improving the robustness and the applicability of data-driven prognostics of aging PEMFC stack and estimating the RUL with limited data. The proposed method is applied to run-to-failure data of PEMFC stack from PHM challenge 2014, which had the life span of 1155 hours. Performances of the approach are judged to encounter parsimony problems. Results show the adaptability of constraint based SW-ELM with limited learning data and its suitability for prognostics of PEMFC stack at frequent intervals.