Dushimepaulette1/Agri-Predict-Rwanda

# Agri-Predict Rwanda: Cassava Leaf Disease Detection **Author:** Dushime Paulette ## About This Project Hi! Welcome to my project, **Agri-Predict Rwanda**. In Rwanda and across Sub-Saharan Africa, cassava is a super important crop that farmers rely on to feed their families and make a living. But, plant diseases like Cassava Mosaic Disease (CMD) can wipe out a whole farm. Since many farmers live in rural villages, it's hard for them to find an agricultural expert to look at their sick plants. My goal is to solve this problem using Artificial Intelligence. I am building a tool where a farmer can just take a picture of a sick leaf with their phone, and the AI will tell them exactly what disease it is! ## How I Built It For this project, I used a public dataset from Kaggle containing over 21,000 pictures of cassava leaves. There are 5 categories (1 healthy class and 4 disease classes). To find the best AI for the job, I ran **8 different coding experiments**: 1. **Traditional Machine Learning:** I started with older models like Random Forest and SVM. They totally failed because to use them, I had to flatten the images, which destroyed the actual shapes of the spots on the leaves. 2. **Basic Deep Learning:** I built my own Convolutional Neural Networks (CNNs). These were better because they could actually "see" the 2D images, but they struggled to learn without memorizing the data. 3. **Transfer Learning (The Winner!):** I used a powerful, pre-trained model from Google called **Xception**. By giving it larger, high-resolution images (224x224), it was finally able to see the tiny yellow veins of the sick leaves. ## The Results My final "Champion" Xception model achieved **79.18% accuracy**! The biggest challenge I faced was that the dataset was super unbalanced (over 13,000 pictures of CMD, but only 1,000 pictures of the rarest disease). To stop the AI from just guessing the most common disease every time, I used a few cool tricks: * **Data Augmentation:** I wro …