An early leaf disease prediction system is necessary for tomato crop as tomatoes are one of the most widely consumed crops worldwide. Tomato leaf diseases such as leaf mold, leaf spot, yellow leaf curl virus have a direct impact on crop yield and farmer income. It also impacts the quality and availability of food. These leaf diseases if not predicted in the early stages can cause up to 70–100% loss in crop yield, resulting in severe economic losses. AI and deep learning models are used extensively in predicting diseases at early stage. To make an early prediction of leaf disease in the tomato crop, an enhanced deep learning technique, that is, EfficientNetB0 with two stages is used. In the first stage, the EfficientNetB0 model is trained only with new top layers to recognize tomato leaf diseases. In the second stage, some of the deep layers of the model are fine-tuned with a smaller learning rate to get better accuracy. To evaluate the performance of the proposed model, various experimentation are done on the curated tomato leaf data set containing healthy and various diseased samples. The proposed method achieved better accuracy of 99.63% compared to traditional machine learning models. The early detection of tomato crop disease with better accuracy helps the farmers to take timely action to avoid crop damage and also to ensure high productivity and sustainable farming practices.