Lumpy Skin Disease Prediction Based on Meteorological and Geospatial Features using Random Forest Algorithm with Hyperparameter Tuning

Abstract

Lumpy Skin Disease is one of the diseases in cattle that has just entered Indonesia. Early prevention of the spread of the disease is needed. Machine learning helps in classifying LSD by utilizing existing LSD datasets from Mende-ley Data. One of the problems in classification using machine learning is the imbalance of data so that resampling techniques are needed. The purpose of this study is to optimize the Random Forest classification in predicting LSD with Genetic Algorithm (GA) as a hyperparameter tuning and using SMOTE as a resampling technique for unbalanced datasets. The results of the experiments conducted showed that the use of SMOTE and GA in the Random Forest Algorithm gave an increase in the Recall and F1 score value from 0.90 to 0.99 respectively and an increase in AUC scores from 0.94 to 0.98. Accuracy of 0.99 indicates that the model can classify cattle that are not infected properly. The recall metric is a concern in this study on the classification of LSD with the hope that the higher the score, the misclassification of infected cows predicted by the model to be healthy. There are still opportunities to improve recall by using various resampling techniques and determining the right model parameters at the begining.