Due to the rapid advancement of computer vision-based technology, automated disease recognition using vision-based technology has become more popular in recent times. In this paper, we have proposed an optimal solution to identify diseased carrots. Firstly, various image-preprocessing techniques have been performed on the collected images. Then, we have used the k-means clustering technique for segmenting the disease-affected region, and then 10 different Statistical and Gray-Level Co-occurrence Matrix (GLCM) features have been extracted from the segmented images. The Mutual information-based feature selection technique is applied. To handle the class imbalance problem, we have performed the synthetic minority oversampling technique (SMOTE) on the training dataset after selecting the top N (5 <  = N <  = 10) features. After that we have trained 5 different machine learning (ML) classifiers. Then, we have used four performance metrics and plotted Receiver Operating Characteristic (ROC) curve to evaluate the classifiers’ performance using the test dataset. From the result analysis, the Random Forest classifier with the top nine features has outperformed all the applied classifiers with the highest accuracy of 94.17%. Lastly, we have found that the features selection technique has helped to minimize the computational cost.