A Comparative Analysis of Uncertainty Assessment for Annual Yield Prediction of Citrus Growth Using FIS and ANFIS Models

Abstract

Accurate prediction of citrus fruit yield is essential for effective agricultural planning, resource allocation, and decision-making. This study aims to compare the uncertainty analysis of developed Fuzzy Inference System (FIS) and Adaptive Neuro-Fuzzy Inference System (ANFIS) models in the context of predicting the annual yield of citrus growth. To achieve this, a comprehensive dataset comprising relevant features such as climate variables, soil conditions, and historical yield records is collected. FIS and ANFIS models were constructed using average temperature, average rainfall, and average relative humidity as input parameters and annual citrus yield as output parameters for the 1980-2019 harvesting season. Out of 40 historical data sets, 35 of them were used to train the models. The last five years were utilized for testing the proposed models. The proposed FIS and ANFIS models were found to be in close agreement with their actual counterparts, i.e. R2 values were 0.913 and 0.935 for FIS and ANFIS, respectively. To evaluate the uncertainty associated with the predictions of both models, a Monte Carlo simulation technique is employed. Preliminary results indicate that the FIS and ANFIS models exhibit promising performance in predicting the annual yield of citrus growth. However, a detailed comparison of uncertainty metrics suggests that the ANFIS model tends to provide more precise and reliable predictions, with narrower confidence intervals, than the FIS model. This could be attributed to the adaptive learning capabilities of ANFIS, allowing it to effectively capture complex nonlinear relationships between input variables and citrus yield.