MULTI SENSOR DESIGN FRAMEWORK FOR PESTE DES PETITS RUMINANTS GOAT DISEASE DETECTION USING MACHINE LEARNING
Keywords:
Machine Learning; PPR; Peste des Petits Ruminants; Random Forest; XGBoost; SMOTE; Goat Health; Disease Outbreak Algorithm; Goat Disease; KNNAbstract
The outbreaks of diseases in animals have in recent years become a critical area of prediction and prevention as a means of protecting livestock, food security and protecting the rural economies. This report provides a machine learning pipeline to forecast the potential risk of an outbreak of Peste des Petits Ruminants (PPR) using the surveillance data (historical data). The data is a collection of information in several countries, and the variables used are the nature of the disease, the type of animal, the number of outbreaks, vaccinations, and the vulnerable population. The preprocessing stage consisted of cleaning and converting numerical and nominal features. The most important steps were the encoding of the categorical variables, the construction of new features, including mortality rate and vaccination coverage, and the definition of a binary target variable that indicated risk of outbreaks. Visualizations and correlation matrices were used as the exploratory data analysis tools. SMOTE was applied in order to deal with class imbalance in the target variable. All classifiers included in it were trained and assessed by the measures of accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrices using Logistic Regression, Random Forest, XGBoost, SVM, and KNN. SelectKBest and MinMaxScaler are feature selection and scaling methods that were used to enhance performance. RandomizedSearchCV was also used to optimize Random Forest in order to increase the ROC-AUC. It was found that ensemble techniques such as XGBoost and Random Forest are more effective at forecasting outbreak risks than others. The last finalized tuned Random Forest model showed good generalization and it was deployed.


