This article deals with model averaging as an alternative regression technique for high-dimensional data especially in chemometrics where statistical approach is used to extract any information contained in a chemical dataset. Our simulation study indicated that model-averaging (MA) works better in high-correlated data than in low-correlated data. The result also designated MA with weighting procedure based on Mallows’ Cp and Jackknife criteria produce better predictions compared to Akaike information criterion (AIC)-based of weight if the candidate models are constructed by randomly grouping the covariates. Moreover, the prediction performance tent to increase along with the number of variables in a candidate model. We illustrated the methods to regress the concentration of curcuminoid in curcumin specimen as a function of their spectra determined by Fourier Transform Infra-red (FTIR) instrument.