Browsing by Subject ",Convolutional Neural Networks"
Now showing 1 – 1 of 1
  • Item
    Interpretable foot and mouth disease detection
    (Makerere University, 2025) Zirimabagabo, Anslem ; Kawooya, Barry Isaac ; Namuwanga, Aisha
    Foot and Mouth Disease (FMD) is a highly contagious viral infection that hinderslivestock production, resulting in severe economic losses in Uganda[21]. This study investigates an interpretable machine learning approach to FMD detection by using two distinct datasets: a numerical dataset sourced from Uganda for early detection, and an image based dataset collected from the internet for visual diagnosis. Exploratory Data Analysis (EDA) was conducted to assess feature distributions, identify class imbalances, and uncover correlations among epidemiological and environmental factors such as rainfall, temperature, livestock density, and geographic proximity to national parks and borders[5]. A total of six models were developed—four trained on the numerical dataset and two on the image dataset. For early detection, models including Random Forest, Support Vector Machine (SVM), Logistic Regression, and Gradient Boosting were evaluated. The XGBoost model, when combined with the Synthetic Minority Oversampling Technique (SMOTE), achieved the highest accuracy of 82%. However, performance decreased in scenarios involving dynamic or imbalanced data distributions, underscoring the need for adaptive learning strategies.[15] In the image-based classification task, deep learning models comprising a custom Convolutional Neural Network (CNN) and ResNet50 were implemented. Among these, ResNet50 achieved the highest accuracy of 97%, demonstrating strong potential for visual FMD symptom detection. To enhance transparency and model trustworthiness, SHAP was employed to explain feature importance in numerical models, while Grad-CAM was used to generate class activation maps for CNN-based image models[14]. This report emphasizes the value of integrating explainable artificial intelligence (XAI) and adaptive machine learning in livestock disease diagnostics. The proposed approach provides a foundation for developing robust, data-driven decision support systems to strengthen early warning and surveillance mechanisms for FMD in Uganda.