Near-Infrared Spectroscopy Analysis: Extracting Relevant Information to Build Reliable Models

In the field of near-infrared spectroscopy analysis, it is crucial to effectively extract relevant information from complex data and construct reliable models. When selecting training sets, sample representativeness must be carefully considered. Currently, multiple methodologies are available, including Random Sampling (RS), Kennard-Stone (KS), and Sample Set Partitioning based on joint X-Y distances (SPXY). Additionally, other techniques such as clone algorithms, cluster sampling, and genetic algorithms offer alternative approaches. Each method possesses distinct advantages and limitations that must be evaluated according to specific application requirements. For instance, Random Sampling provides straightforward implementation through basic random number generation but may lead to sample imbalance issues. Conversely, the KS algorithm ensures uniform distribution across the sample space by sequentially selecting points farthest from existing selections, though this requires more computational resources through distance matrix calculations. Therefore, selecting an appropriate sample selection method necessitates comprehensive consideration of multiple factors to determine the most suitable approach for the analytical scenario.