An active learning representative subset selection method using net analyte signal.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Accurate spectroscopic calibration models require representative samples.
• Identifying representative samples before measurement is challenging, time-consuming, and expensive.

Purpose of the Study:

• To develop an efficient method for selecting representative samples for calibration sets.
• To reduce the cost and time associated with reference measurements in spectroscopic analysis.

Main Methods:

• Utilized the Euclidean norm of the net analyte signal (NAS) vector to assess sample representativeness.
• Computed projection matrices, NAS vectors, and scalar values from initial sample data.
• Sequentially selected candidate samples based on the largest distance between NAS vectors of candidate and existing sets.

Main Results:

• The proposed method demonstrated higher efficiency compared to random sample selection.
• Significantly reduced the time and financial resources required for reference measurements.
• Enabled the creation of more robust and accurate calibration models with fewer samples.

Conclusions:

• The NAS vector-based method provides an effective strategy for optimizing calibration set selection.
• This approach leads to substantial savings in resources while maintaining predictive accuracy.
• Offers a valuable tool for improving the efficiency of spectroscopic calibration development.