Chemiresistive Sensor Array and Machine Learning Classification of Food.

Area of Science:

• Materials Science
• Analytical Chemistry
• Machine Learning

Background:

• Identifying complex odors with sensor arrays is difficult.
• Carbon nanotube (CNT) sensors offer potential for chemical sensing applications.

Purpose of the Study:

• To develop a robust method for category-specific odor classification using a CNT sensor array.
• To identify optimal sensor subsets for differentiating between cheese, liquor, and edible oil odors.

Main Methods:

• Utilized a two-stage machine learning approach with an array of 20 CNT-based chemical sensors.
• Employed independent selector classification accuracy and combinatorial scanning to find optimal sensor subsets.
• Applied k-nearest neighbors and random forest models for multiclass-time series classification.

Main Results:

• Achieved high classification accuracy for cheese (91%) and liquor (78%) samples in independent test sets.
• Demonstrated 73% accuracy for edible oil classification using the optimized sensor subsets and machine learning models.
• Successfully differentiated between distinct odor categories based on sensor array responses.

Conclusions:

• The proposed machine learning protocol enables effective category-specific odor identification using CNT sensor arrays.
• Optimized sensor subsets enhance classification performance for complex odor mixtures.
• This methodology shows promise for practical applications in food quality control and environmental monitoring.