Evaluation of Semantic XAI for Soil Moisture Prediction: Insights from Tunisian Farmers

The dataset comprises evaluation results from a survey conducted with Tunisian Farmers' Union to assess different explanation styles: human, raw LLM, and semantic LLM in a soil moisture prediction context. The collected data is organized into a compressed folder containing three subfolders, each corresponding to a separate evaluation instance (ID1, ID2, and ID3). Within each instance folder, there are two CSV files: Ranking.csv and Metrics_Evaluation.csv. The Ranking.csv file captures the farmers' rankings of the three explanation styles based on their overall preferences. The Metrics_Evaluation.csv file contains farmers' ratings of each explanation style on a scale from 1 to 4 across four evaluation metrics: Clarity, Relevance, Usefulness, and Trustworthiness. A rating of 1 represents the highest level of Clarity, whereas for Relevance, Usefulness, and Trustworthiness, a rating of 1 indicates the lowest score. This dataset provides insights into how farmers perceive different explanation styles in terms of their clarity, applicability, and reliability for agricultural decision-making.