Precision Agriculture: Improving Water Stress Forecasting in Tomato Plants Using Bioristor Data and Advanced ML Techniques

The water stress in crops, especially tomato plants, is highly challenging to yields and qualities. This study involves improving the forecasting of water stress by using Bioristor sensor information combined with advanced machine learning models. Bioristor sensors capture ion concentration changes in plant saps; hence very crucial real-time data. We developed models such as RNN, LSTM, Bi-GRU, and ARIMA, for predictions and time series forecasting. We applied the models on classification with Decision Trees as well as Random Forest. The performance of the model was measured using applications in accuracy, precision, recall, and confusion matrix analysis metrics, when predicting 1 and 24 hours in advance, using classifications of 2-state (stressed vs. not stressed) and 4-state (healthy, stressed, uncertain, recovery). Our results suggest promising accuracy for the prediction of water-stress conditions, offering a framework which may be applied in real-world precision agriculture to optimize strategies for water management.