This paper concerns the effect of the subsampling ratio on the subagging approach for multiple linear regression with variable selection by the successive projections algorithm. Investigations involving simulated data, as well as near-infrared spectrometric determination of moisture and protein in wheat and distillation temperatures (T10 and T90), specific mass and sulphur in diesel, are presented. In terms of prediction ability and sensitivity to noise, the best results were obtained for subsampling ratios around 40%.
multivariate calibration; successive projections algorithm; subagging; near-infrared spectrometry; wheat; diesel
