Chaos-Embedded Multi-Objective Intelligent Optimization-Based Explainable Classification Model for Determining Cherry Fruit Fly Infestation Levels Using Pomological Data

The European cherry fruit fly (<i>Rhagoletis cerasi</i> L.) poses a significant pest threat to cherry production due to its rapid reproduction and host specificity, causing substantial economic damage. This study presents a novel, explainable, and biologically inspired data-driven classification model based on fruit characteristics to support targeted and sustainable pest control strategies. In research conducted at four different locations in Elazığ province, three population classes were determined based on the number of adult individuals caught in traps, and 10 different fruit characteristics were measured in fruit samples belonging to each class. The data used in this study are original data obtained by the authors. To examine the relationship between pomological characteristics of cherry fruit and cherry fruit fly density, the Chaotic Rule-based-Strength Pareto Evolutionary Algorithm2 (CRb-SPEA2) method, developed as a multi-objective and chaos-integrated evolutionary rule mining framework, was adapted. The developed algorithm aimed for high performance, interpretability, and transparency. Accuracy, Precision, and Recall metrics, which are conflicting objectives, were optimized with Pareto-optimal solutions, yielding selectable results for domain experts. To increase population diversity and reduce the risk of early convergence and getting stuck in a local optimum, the Tent chaotic mapping mechanism was also integrated into the system. Furthermore, the model was trained without the need for predefined automatic discretization of the continuous value ranges of the attributes. The proposed model achieved superior results across all classes, with the highest accuracy rate of 82.6% recorded in the High class, demonstrating excellent sensitivity and recall values.