In this paper (purchase required) the authors build a Raman spectroscopy classification model to differentiate 5 brands of very similar dairy products. The model was trained on 30 retail samples of each brand. The authors reported significantly improved feature selection and model performance (particularly in detecting differences in dairy fat content, the key question for adulteration) when they used a data fusion approach of 4 different Machine Learning protocols rather than training using one protocol. They recommend this fusion approach for constructing classification models.
Comments