A Practical Method to Estimate the Resolving Power of a Chemical Sensor Array: Application to Feature Selection

Area of Science:

• Analytical Chemistry
• Sensor Technology
• Information Theory

Background:

• Calculating analytical figures of merit for sensor arrays with low selectivity is challenging.
• Existing methods often fail with non-linear sensors or unknown sensitivity matrices.

Purpose of the Study:

• To present a practical methodology for estimating the Resolving Power of sensory systems.
• To address limitations in current methods for evaluating sensor array performance.

Main Methods:

• Adapted Shannon's information theory and Gardner/Barlett's chemical sensor adaptation.
• Combined dimensionality reduction with convex hull algorithms to estimate data volume.
• Validated with synthetic and real-world data from temperature-modulated MOX gas sensors.

Main Results:

• Successfully estimated Resolving Power for chemical sensor arrays with non-linear sensors and heteroscedastic noise.
• Method does not require intrinsic dimensionality to be less than input dimensionality.
• Circumvents the need for an unknown sensitivity matrix.

Conclusions:

• Provides a novel and practical figure of merit for benchmarking broad-response gas sensor arrays.
• The methodology is robust and applicable to real-world sensor systems.
• Fills a critical gap in the literature for sensor array performance evaluation.