<![CDATA[A hybrid artificial intelligence (AI) system was successfully developed in this study, combining the ResNet50 architecture as an image feature identifier and the XGBoost algorithm for final classification. This model was used to detect six disease variations using 5,505 wheat leaf photographs. To ensure model stability, rigorous testing was conducted using two methods: Stratified 5-Fold Cross-Validation on the entire data set and independent testing using 300 images (equally divided into 50 samples per class).The test results demonstrated very solid performance. The model recorded an average global accuracy of 94.66% (±0.21%) using the K-Fold method, and an accuracy of 94.67% and a Macro F1-Score of 0.9445 in the independent testing. Through confusion matrix mapping, the model successfully classified the Healthy, Black Rust, and Septoria categories perfectly (a score of 1.00). However, there was still a minor error in the case of eight samples being confused between Brown Rust and Yellow Rust due to the visual similarity of the orange-yellowish coloration early in the infection period. Furthermore, the Feature Importance assessment demonstrated that the XGBoost decision base is transparent (Explainable AI). This AI accurately focuses on clinical signs of plants such as chlorosis symptoms and spot texture, while ignoring background objects such as weeds and soil. This combination of methods creates a stable, efficient system with a response time of only 18.5 milliseconds per photo, and a biologically valid decision base.]]>
