Comparison of various classification techniques for supervision of milk processing

Area of Science:

• Food Science
• Analytical Chemistry
• Process Engineering

Background:

• Maintaining milk quality and safety is paramount in the dairy industry.
• Identifying process anomalies like temperature fluctuations, fat variations, or adulteration is essential for quality assurance.

Purpose of the Study:

• To develop and evaluate a method for detecting anomalies during milk processing.
• To assess the efficacy of Raman spectroscopy combined with chemometric methods for anomaly detection.

Main Methods:

• Utilized Raman spectroscopy to analyze milk samples.
• Employed classification algorithms: linear discriminant analysis (LDA), decision tree (DT), support vector machine (SVM), and k-nearest neighbor (KNN).
• Tested the ability of these methods to detect anomalies including added water, cleaning solutions, fat variations, and temperature changes.

Main Results:

• Linear discriminant analysis and decision trees showed limited success in anomaly classification.
• K-nearest neighbor (84.8% accuracy) and support vector machine (81.4% accuracy) demonstrated promising results in classifying different sample groups.
• Both KNN and SVM proved capable of differentiating between normal and anomalous milk samples.

Conclusions:

• Raman spectroscopy coupled with KNN and SVM algorithms offers a viable approach for detecting anomalies in milk processing.
• These methods can assist the dairy industry in identifying and rectifying process deviations, ensuring product quality and safety.