Potatoes hold significant prominence as a primary tuber crop on a global scale, being cultivated in more than 125 nations. Potato, following rice and wheat, is a staple crop that is consumed daily by around one billion individuals across the globe. Nevertheless, the potato crop is seeing a decline in both its quality and quantity as a rsult of various fungal and bacterial diseases. The challenge of early disease identification arises from variations in climatic circumstances, plant species, and the manifestation of plant disease symptoms. Numerous machine learning methods have been developed in recent research endeavors to accurately identify and classify potato leaf diseases. This study mainly deals with the Resnet50ViT model for classifying diseases from potato leaves. The Resnet50ViT models contain pre-trained Resnet50 and ViT, which makes them superior to all models, including Resnet50, ViT, Swin-T, and CSwin-T. It performs the classification of potato leaves into three classes, early blight, healthy blight, and late blight. According to the Multi-Criteria Decision Analysis (MCDA) approach TOPSIS, Resnet50ViT exhibits superior performance compared to other models, as evidenced by its score of 1. Furthermore, due to its cost-effectiveness, our proposition is very suitable for the screening of potato leaves.