Explainable Classification of Different Coloration Stages in Cherry Fruits Using a Hybrid RF-ACO Model Based on Pomological and Cherry Fly Data

The coloring process in cherry fruit affects both fruit quality and the cherry fly (Rhagoletis cerasi). This is critically important for host preferences of major pests such as cherry fly. In this study, a randomized classification method was used to classify five distinct coloration stages of cherry fruit based on pomological characteristics and cherry fly density data. An explainable hybrid model combining Random Forest (RF) and Ant Colony Optimization (ACO) algorithms has been developed. In this study, fruit samples of the Ziraat 900 variety were collected from four different cherry orchards in Elazığ province during five different coloration stages. Pomological characteristics such as weight, width, length, height, stem length, firmness, seed weight, soluble solids content (SSC), NaOH, and acidity were determined, and adult cherry fly densities were also recorded. In the proposed method, candidate decision rules generated by the RF algorithm were optimized using the ACO algorithm, and the most distinctive rule sets were selected. The results showed that the proposed RF-ACO model achieved a 99.48% accuracy rate and exhibited higher performance than many common machine learning methods. Feature significance analysis revealed that SSC, cherry fly density, NaOH, and acidity were the most effective parameters in the classification process. The model not only provided high accuracy thanks to the explainable IF-THEN rules it generated, but also allowed for expert interpretation of the decision-making process. The findings offer significant contributions to the development of decision support systems for determining cherry ripening periods and controlling the cherry fruit fly.