Prediction of the purity of stingless bee honey using fluorescence-based UV-visible spectrum data with a generative model approach

Abstract

Abstract The stingless bee has limitations in producing honey, the price of the honeycomb is relatively more expensive, and the characteristic of honey is that it has a high water content, which affects the short shelf life. According to information obtained from cultivators in several different locations, managing this type of honey also has a number of challenges. Starting from weather factors that influence food/vegetation conditions, limited markets, and to predators that are always lurking. From these several things, the opportunity for the presence of innovation on the downstream side could help significantly in meeting the need for stingless honey products whose purity is maintained. Several studies in the field of spectroscopy show that some of the algorithms used are discriminant model approaches. One of the problems of using discriminant models that is observed in this research is the overlapping data distribution, high outliers, and the accuracy and performance values of the model can still be improved. A generative model approach with Expectation Maximization (EM) algorithm for this research shows the results of increasing the accuracy value from 89% to 94%, and the model performance shows from 2.8 seconds to 0.5 seconds.