Yield prediction in apples using Fuzzy Cognitive Map learning approach

This work investigates the yield modeling and prediction process in apples (cv. Red Chief) using the dynamic influence graph of Fuzzy Cognitive Maps (FCMs). FCMs are ideal causal cognition tools for modeling and simulating dynamic systems. They gained momentum due to their simplicity, flexibility to model design, adaptability to different situations, and easiness of use. In general, they model the behavior of a complex system, have inference capabilities and can be used to predict new knowledge. In this work, a data driven non-linear FCM learning approach was chosen to categorize yield in apples, where very few decision making techniques were investigated. Through the proposed methodology, FCMs were designed and developed to represent experts’ knowledge for yield prediction and crop management. The developed FCM model consists of nodes linked by directed edges, where the nodes represent the main soil factors affecting yield, [such as soil texture (clay and sand content), soil electrical conductivity (EC), potassium (K), phosphorus (P), organic matter (OM), calcium (Ca) and zinc (Zn) contents], and the directed edges show the cause-effect (weighted) relationships between the soil properties and yield. The main purpose of this study was to classify apple yield using an efficient FCM learning algorithm, the non-linear Hebbian learning, and to compare it with the conventional FCM tool and benchmark machine learning algorithms. All algorithms have been implemented in the same data set of 56 cases measured in 2005 in an apple orchard located in central Greece. The analysis showed the superiority of the FCM learning approach in yield prediction.

► A Fuzzy Cognitive Map – based approach for yield prediction in apples was proposed. ► A data driven learning algorithm for Fuzzy Cognitive Maps was applied for classification. ► Hebbian-based FCM learning approach was chosen to categorize yield in apples. ► Data driven FCM learning approach was compared with benchmark machine learning algorithms using WEKA toolbox.