Interpretability of artificial intelligence models that use data fusion to predict yield in aeroponics

There is an increasing demand for healthy and fresh foods, and predicting yield effectively is important to improve production, especially in methods like aeroponics. This paper has two main goals: (i) use data fusion to improve yield prediction in aeroponics, and (ii) find which features are more relevant for yield prediction of six different crops. To reach these goals, a number of artificial intelligence models and an interpretability analysis based on SHapley Additive exPlanations (SHAP) have been implemented. The models were trained using 200 samples that were collected in a nine-month period, including information from different air and water quality sensors in addition to manually recorded data, reaching in the end a coefficient of determination value $$R^{2}$$ = 0.752 for the validation dataset in the best case (CNN-based model). As a result, two main features were identified in the dataset: Room $$CO_{2}$$ and Reservoir Temperature, along with other useful insights of how these features influence predictions. SHAP values also provided important information for feature selection. These results could be the first steps towards the full automation of an aeroponics crop production system.