Guava (Psidium guajava) is an important part of tropical agriculture with its high nutritional value and economic return. However, diseases in this fruit cause serious losses in productivity and product quality. The manual and subjective nature of traditional diagnostic methods is insufficient to meet the speed and accuracy requirements of modern agriculture. Therefore, the aim of the study is to identify guava fruit using a model that combines ConvNeXt and machine learning. In this study, ConvNeXt-Tiny, ConvNeXt-Small, ConvNeXt-Base, ConvNeXt-Large and ConvNeXt-XLarge architectures were used as feature extractors. The features extracted from these variants were classified using Support Vector Machines (SVM), Random Forest (RF) and Logistic Regression (LR) algorithms. In addition, different numbers of features (50, 100 and 200) were selected from each ConvNeXt variant by SelectKBest method and new feature vectors were created by combining these features. The classification performances of these feature vectors were also evaluated with SVM, RF, and LR. According to the experimental results, the ConvNeXt-XLarge-SVM model achieved an accuracy of 0.997. The CS200-ConvNeXt-SVM model after feature selection and feature fusion achieved 1.000 in all performance metrics. Contextual Importance and Utility (CIU), one of the explainable artificial intelligence methods, was used to identify the features that contribute to the decision mechanism of the proposed model. The contributions of the features determined by the CIU method for the CS200-ConvNeXt-SVM model were analyzed on a class basis, and the top 10 features with the highest contribution were identified for each class. It was found that the F570 feature contributed the most in the Healthy guava class, the F625 feature contributed the most in the Anthracnose class, and the F626 feature contributed the most in the Fruit fly class. In addition, evaluations were conducted across different datasets to assess the generalization performance of the proposed model. The findings suggest that combining ConvNeXt variants and machine learning algorithms is an effective approach for guava fruit disease classification.