Sensor Failure Tolerable Machine Learning-Based Food Quality Prediction Model.

Area of Science:

• Agricultural Science
• Sensor Technology
• Machine Learning

Background:

• Food quality assessment is crucial for human health and economic value in agriculture.
• Electronic noses (e-noses) simulate smell using sensors to detect odor compounds for quality assessment.
• Sensor failures in e-noses can compromise accurate food quality evaluation.

Purpose of the Study:

• To propose a machine learning-based failure tolerance strategy for e-nose sensor systems.
• To develop a method that ignores data from failed sensors instead of correcting it.
• To enhance the reliability and accuracy of food quality assessment despite sensor malfunctions.

Main Methods:

• A Single Plurality Voting System (SPVS) classification approach is proposed for failure tolerance.
• Individual classifiers (kNN, Decision Tree, LDA) are trained for each sensor feature.
• A composite classifier is built based on the outcomes of individual classifiers.

Main Results:

• The SPVS approach effectively tolerates sensor failures by ignoring problematic data.
• The proposed method maintains acceptable prediction accuracy even with sensor malfunctions.
• Promising results were demonstrated using a case study on beef cut quality assessment.

Conclusions:

• The developed machine learning-based failure tolerance method enhances food quality assessment reliability.
• Ignoring failed sensors offers an advantageous alternative to traditional data correction techniques.
• The approach shows significant potential for improving e-nose applications in food quality control.