Vitoria pineapple yield predictions by neuro-fuzzy modeling and linear regression

Abstract

Hybrid intelligent systems that combine artificial intelligence techniques, such as neural networks and fuzzy logic, have become common for the development of complex models to predict and estimate variable parameters. The objective of this study was to compare predictions of Vitoria pineapple yields by Adaptive-Network-Based Fuzzy Inference Systems (ANFIS) and linear or quadratic regression models. The prediction models developed calculate the fruit fresh weight based on the D leaf fresh weight (DLFW) and stem diameter (SD), measured at the time of floral induction. ANFIS were developed using the genfisOptions function of the Neuro Fuzzy Designer toolbox of the Matlab program (Mathworks®- Neuro Fuzzy Designer, R2018a), considering DLFW and SD as the entry parameters, single and combined. The yield prediction error was calculated using the root mean square error (RMSE). The RMSE found for all ANFIS developed were lower than that predicted by linear or quadratic regression models. The lowest RMSE was obtained when the parameters DLFW and SD were combined for the development of the ANFIS. Therefore, the results showed that the use of neuro-fuzzy modeling (ANFIS) for predicting Vitoria pineapple yield presents better results than the use of linear or quadratic regression models.