A machine learning-based electronic nose system using numerous low-cost gas sensors for real-time alcoholic beverage...

Area of Science:

• Analytical Chemistry
• Sensor Technology
• Machine Learning

Background:

• Dogs' superior olfaction highlights the potential of advanced odor classification systems.
• Homogeneous metal-oxide semiconductor (MOS) sensors, despite poor selectivity, can yield distinctive patterns for classification.
• Leveraging numerous low-cost sensors offers a viable alternative to expensive, complex systems.

Purpose of the Study:

• To develop a real-time alcoholic beverage classification system using numerous low-cost gas sensors and machine learning.
• To investigate the influence of sensor quantity and data preprocessing on classification accuracy.
• To evaluate the performance of the proposed electronic nose system against commercial standards.

Main Methods:

• An electronic nose system was constructed using 30 homogeneous MOS gas sensors.
• Alcoholic beverage samples (whiskey, soju, white wine) were analyzed in a gas chamber.
• Time-series sensor data were preprocessed into four datasets and analyzed using machine learning algorithms, including a pretrained Linear Discriminant Analysis (LDA) model.

Main Results:

• The system achieved high classification accuracy, exceeding 99%, with optimal performance at 99.83 ± 0.21% when using more sensors.
• Classification performance was significantly influenced by the data preprocessing techniques employed.
• The proposed system demonstrated classification performance comparable to commercial electronic nose systems.

Conclusions:

• A low-cost, high-accuracy electronic nose system for real-time alcoholic beverage classification is feasible.
• The number of gas sensors is a critical factor for improving prediction accuracy.
• The study validates the use of machine learning with heterogeneous sensor responses for effective odor analysis.