5. Conclusion
This paper described a novel forward selection algorithm SPA for variable selection in multivariate calibration. SPA employs simple operations in a vector space to obtain subsets of variables with small collinearity. It is shown to demand a smaller computational workload than guided random search techniques GRSA , such as genetic algorithms GA , mainly when the total number of variables gets large. Also, unlike SPA, the stochastic nature of GRSA makes them non-dependable for finite optimization time. The restriction in the number of wavelengths to be selected which cannot be larger than the number of . calibration samples is a limitation of SPA. How- ever, this was not a major handicap in the present application. It can also be argued that, if many spectral variables are needed to discriminate the analytes, then a large number of samples will also be required to perform the calibration. The use of SPA with data sets gathered by different multicomponent instrumental techniques is being investigated. Future research could also attempt to employ SPA to provide a initial solution to be further refined by GA. Preliminary results have shown that such procedure allows GA to approach the optimum selection in a smaller time and also help alleviate the uncertainty in its performance