Incorporating active learning into machine learning techniques for sensory evaluation of food

Abstract

The sensory evaluation of food quality using a machine learning approach provides a means of measuring the quality of food products. Thus, this type of evaluation may assist in improving the composition of foods and encouraging the development of new food products. However, human intervention has been often required in order to obtain labeled data for training machine learning models used in the evaluation process, which is time-consuming and costly. This paper aims at incorporating active learning into machine learning techniques to overcome this obstacle for sensory evaluation task. In particular, three algorithms are developed for sensory evaluation of wine quality. The first algorithm called Uncertainty Model (UCM) employs an uncertainty sampling approach, while the second algorithm called Combined Model (CBM) combines support vector machine with Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, and both of which are aimed at selecting the most informative samples from a large dataset for labeling during the training process so as to enhance the performance of the classification models. The third algorithm called Noisy Model (NSM) is then proposed to deal with the noisy labels during the learning process. The empirical results showed that these algorithms can achieve higher accuracies in this classification task. Furthermore, they can be applied to optimize food ingredients and the consumer acceptance in real markets.