Data-Driven Approach toward the Quantification of Gases in a Complex Mixture Using a Non-Selective Single Metal Oxide...

Area of Science:

• Materials Science
• Sensor Technology
• Artificial Intelligence

Background:

• Metal oxide semiconductor (MOS) sensors offer high sensitivity for volatile organic compound (VOC) detection.
• However, poor selectivity limits their performance in real-world applications like diagnostics and air quality control.

Purpose of the Study:

• To develop a machine learning (ML) framework for a single, non-selective MOS sensor.
• To achieve accurate VOC classification and concentration prediction, overcoming selectivity limitations.

Main Methods:

• Utilized RF-sputtered nickel oxide thin film with gold contacts as the MOS sensor.
• Evaluated time-independent and time-dependent features using ensemble methods, ANNs, LSTMs, and GRUs.
• Applied regression analysis for concentration prediction.

Main Results:

• Time-independent features with ensemble models achieved 98% accuracy for baseline classification.
• Time-dependent features with sequential models reached >94% accuracy by capturing adsorption-desorption kinetics.
• Regression techniques improved predictive capabilities, showing higher R² and lower RMSE.

Conclusions:

• ML-based analysis complements material innovation to enhance MOS sensor selectivity for VOCs.
• This approach enables scalable, real-time monitoring in complex gas environments.
• The framework is adaptable for detecting other toxic gases, pollutants, and biomarkers.