Neural Network Model for Predicting Apple Yield Based on Arrival of Phenological Stage in Conjunction with Leaf disease, Soil and Weather Parameters

Forecasting the yield of a vital crop-like apples needs to be done in a way that is both accurate and dependable to facilitate effective planning and management. There have been attempts made to forecast apple production, the majority of which have involved the utilisation of statistical methods that contain a restricted number of indicator factors or fruit counting techniques which required technical expertise and advance equipments. The suggested method utilises a neural network (NN) to forecast the yield of apple crops in an orchard. This prediction is made by considering many factors, such as the identification of phenological stage, as well as the timing of their arrival. In addition, the system takes into account soil, weather and leaf disease characteristics, which interact with the aforementioned factors. The proposed approach aims to estimate crop production across five distinct classes throughout various phenological phases. This prediction would consider factors, such as the timing of phenological stage emergence (early emergence, regular emergence, or late emergence), soil parameters, weather-related parameters, and presence or absence of leaf disease. The proposed yield prediction model shows a f1-score of 0.91, precision of 0.91, recall of 0.92, and accuracy of 92%. It is compared with other popular classical machine learning (ML) algorithms, such as Random Forest, Decision Tree, and Gradient Boosting. The challenge lies in the process of automatically predicting yields in orchards. Despite the considerable efforts dedicated to the development of automated techniques for yield estimation, the prevailing methods employed at now rely on fruit counting, which proves to be effective just within a narrow timeframe of 1–4 weeks before to harvest. This model will assist agricultural producers in making timely decisions on the implementation of contingency plans in the event of subpar or minimal yields.