IMPLEMENTATION OF RANDOM FOREST FOR ANIMAL PROTEIN CLASSIFICATION THROUGH HYPERPARAMETER OPTIMIZATION

Accurate identification of animal protein types is crucial to ensure food authenticity and safety, particularly in the context of compliance with halal principles. This study aims to implement the Random Forest (RF) algorithm to classify four types of animal protein—broiler chicken, free-range chicken, pork, and beef through hyperparameter optimization using GridSearchCV. The dataset was evaluated using 5-fold cross-validation, and feature importance analysis was conducted to identify the variables that contributed most to classification. Results showed that RF with optimized hyperparameters achieved a test accuracy of 92.81%, with macro-average precision, recall, and F1-score of 93%. The model performed best for the broiler chicken and pork classes, while the beef class exhibited a higher misclassification rate, likely due to the similarity of spectral characteristics among classes. ODOR, CO₂, H₂, NH₃, and VOC were identified as the key indicators for distinguishing animal protein types. This study contributes to halal authentication by integrating FTIR spectral data with optimized Random Forest, enabling efficient and accurate classification. Although RF proved reliable and capable of handling high-dimensional data, the study is limited by dataset size and spectral feature complexity. Future research is recommended to explore deep learning architectures, such as Convolutional Neural Networks (CNN), with larger FTIR datasets to improve model generalization and robustness