In modern agriculture, controlled environment techniques like hydroponic greenhouses are becoming increasingly popular for their efficiency and higher crop yields. This paper presents a new method for predicting temperature and humidity in smart hydroponic greenhouses, combining neural network and granular computing. We have used a Feed-Forward Neural Network and fuzzy sets to classify greenhouse climates based on thresholds derived from tomato plant cultivation. Additionally, our solution improves system credibility and reliability by
identifying Out-Of-Knowledge predictions. Moreover, our system reduces neural network complexity while maintaining accuracy, making it suitable for low-power devices and edge/extreme-edge implementation. This work aims to improve the interpretability of neural network results, advancing practical applications in agriculture.