| Summary: | Exposure to pollutants like ozone and nitrogen dioxide gas can cause serious health issues and harm the environment. Therefore, the interest in air quality and its impact on health and well-being has been steadily increasing over the years, making low-cost gas sensing devices combined with artificial intelligence (AI) increasingly popular due to their flexibility and small form factor. While AI provides state-of-the-art performance, it makes the system less transparent and more difficult to trust its decisions. With the aid of three different approaches, this paper seeks to understand and explain the predictions made by complex models for gas sensors. The use of such techniques can increase our confidence in the AI systems embedded in our products in terms of fairness, or impartiality, and robustness, or reliability. This also improves our understanding of sensor behavior and provides a more robust explanation for algorithmic choices.
|
|---|