Objective: To develop a machine learning model integrated with Raman spectroscopy for the rapid quantification of potassium acetate in preservation solutions.
Methods: The study was conducted using a LabRam Soleil Raman spectrometer (532 nm) and a dataset comprising 240 Raman spectra obtained from potassium acetate/glycerol samples. The data were divided into training, validation, and test sets. Five-fold cross-validation and Grid Search were applied for hyperparameter optimization. Four models (MLR, PCR, PLS, and ANN) were developed and evaluated using R², RMSE, and MAPE. Model performance was further compared with potentiometric titration using 10 real-world samples.
Results: The PLS model with six latent variables achieved the best overall performance (R² = 0.9991, RMSE = 0.52%, and MAPE = 1.24%), with a training time of 2.5 seconds and high interpretability. The automated workflow reduced analysis time by 85% and enabled the processing of 96 samples per day. A strong correlation was observed between the PLS-Raman method and potentiometric titration (r = 0.985, p < 0.001), demonstrating high specificity for potassium acetate determination.
Conclusion: The Raman-PLS system provides a rapid, accurate, and reliable analytical tool with strong potential for quality control of biological sample preservation solutions.