Determination of Pure Animal Fat Concentrations and Halal Category from Mixed Animal Fats Raman Spectra

Analysis of chemical and biological compounds generally uses the Raman spectrum because of its non-destructive nature. Raman spectra can be analyzed computationally using chemometric methods. The principal component analysis (PCA) method can be combined with another regression or classification model to determine the concentrations of fat and halal categories from animal fats data. This research aims to determine the best regression methods for finding pure spectrum concentrations in mixed spectra of animal fats based on the lowest root mean square error (RMSE) and mean absolute error (MAE) of the regressor, and to determine the best classifier methods for finding halal categories of animal fats. It is shown that the PCA with decision tree regressor performs better on training data than random forest or K-nearest neighbor to predict pure animal fats concentrations, with mean squared error (MSE) of 0.001 and MAE of 0.002 on training data. The decision tree classifier also performs better than the random forest or K-nearest neighbor classifier in predicting the halal category, with 0.94 recall in the haram class and 0.90 recall in the halal class. However, this model still does not meet the conditions to have 1 recall value in the haram class as expected for real-world applications.