<![CDATA[Early detection of crop disease is an important part of modern agriculture since early detection would help in reducing crop loss and improving food security. The purpose of this study is to develop and evaluate lightweight deep-learning models for disease detection using simulation-based data where the output device would be a mobile device. Training and testing three types of machine learning models, Random Forest (RF), Support Vector Machine (SVM), and Convolutional Neural Network (CNN) on simulated agricultural data of soil health, weather conditions and plant health is a part of the research methodology. To evaluate the models, the accuracy, F1-score and inference time were used. And results indicate that RF and SVM both performed with 100% accuracy (F1 score equal 1.0) whereas the CNN model has 87.5 % accuracy and loss = 0.2279. The CNN model, although it has slightly lower performance, is promising for deployment on mobile as it offers better results. The study concludes that there is room for light-weight CNN models for real-time disease detection on mobile devices. The future study will analyze how CNN architecture can be optimized using real-world data. This study has practical implications for mobile-based solutions for crop disease management in resource-constrained environments. A major weakness is that the data used is simulated data and may not account for the realities of agricultural conditions.]]>
