Selective Sensing of Volatile Organic Compounds Using an Electrostatically Formed Nanowire Sensor Based on Automatic...

Area of Science:

• Sensor Technology
• Machine Learning
• Internet of Things (IoT)

Background:

• Electrostatically formed nanowire (EFN) gas sensors offer ultralow power consumption and compatibility with CMOS technology for mass production.
• Achieving selective gas detection with EFN sensors necessitates advanced identification methods like machine learning.

Purpose of the Study:

• To introduce automatic learning technology for sorting and applying common algorithms to EFN gas sensors.
• To evaluate and compare the performance of tree-based machine learning models for EFN gas sensor selectivity.
• To enhance algorithm accuracy through ensemble methods and analyze feature importance.

Main Methods:

• Implementation of automatic learning to sort and apply machine learning algorithms.
• Discussion of advantages and disadvantages of four top tree-based models.
• Ensemble of unilateral training models to improve algorithm accuracy.
• Analysis of experimental data from two groups of EFN gas sensor tests.

Main Results:

• The CatBoost algorithm exhibited the highest evaluation index among the tested models.
• Feature importance analysis provided insights into the physical meaning of EFN dimensions for classification.
• The study successfully improved the accuracy of gas identification for EFN sensors.

Conclusions:

• CatBoost is a highly effective algorithm for enhancing the selectivity of EFN gas sensors.
• Understanding feature importance aids in model fusion and exploring underlying physical mechanisms.
• This work paves the way for more accurate and selective gas sensing in IoT devices.