Loading...
Publication
Design principles for explainable ai in smart farming: insights from Agronomists' perceptions
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.89 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
This dissertation investigates the development of interfaces for presenting explanations generated by Explainable Artificial Intelligence (XAI) techniques, with the goal of maximizing usability and users’ understanding through the application of Human Computer Interaction (HCI) principles. The work introduces a proof of concept developed in partnership with Associação de Viticultores do Concelho de Palmela (AVIPE), demonstrating an intelligent decision-support system for vineyard disease detection. The development followed a user-centered methodology, starting with low-fidelity prototypes and evolving into a functional prototype integrated into the AgriUXE platform.
This iterative process incorporated co-design sessions to ensure that the solution addressed the real needs of agronomists. A user study was conducted to evaluate the effectiveness of three explanation modalities: a control scenario (diagnosis only), visual explanations using heatmaps, and examplebased explanations. The evaluation considered metrics such as trust, cognitive load, and satisfaction. The results show a clear preference for example-based explanations, which were perceived as more intuitive, direct, and effective for calibrating user trust. This modality enabled 33% of participants to correctly identify an incorrect model prediction, which is a result not achieved by the other scenarios.
Qualitative analysis, based on co-design sessions and the think-aloud protocol, identified key requirements for integrating XAI explanations into the agronomists’ workflow.
The findings indicate that professionals value tools that operate as a “second opinion,” supporting active investigation rather than providing definitive answers. Despite limitations related to sample size and controlled study conditions, the results allowed the formulation of practical guidelines for developing tools that support decisionmaking in agricultural contexts, contributing to a more informed and effective adoption of XAI solutions in smart agriculture. The findings informed a set of design principles for creating usable and trustworthy XAI-enabled interfaces for smart farming. Future work should expand evaluation to more diverse agricultural contexts and assess how explanations influence diagnostic accuracy in real field conditions.
Description
Keywords
Agricultura digital XAI HCI Suporte à decisão agrícola Usabilidade Design thinking Agronomia