Predicting consumer acceptance of packaged meat using L1-regularized ordinal regression
Communications in Agricultural and Applied Biological Sciences - 2015
Abstract
Modified atmosphere packaging (MAP) is a technique that is commonly used to extend the shelf-life of fresh meat. Unfortunately, MAP packages prevent consumers from judging the freshness of the packaged meat based on its smell. Recent research efforts have led to the development of (prototypes of) optical sensors that allow to measure the concentration of multiple volatile organic compounds (VOCs) in the headspace of a package in a non-destructive manner. These measurements can be used to assess the perceived freshness of the meat in that package. To make such an assessment, a model (or relationship) is needed to predict the consumer appreciation of the meat in a package with given VOC concentrations. In this work, we present a machine learning strategy that can be used to learn such a relationship from data. Interestingly, the available data to learn this relationship have several properties that complicate the learning process. For example: consumer appreciation (the response variable) is measured on an ordinal scale, appreciation is personal and not consistent over consumers, the input is high-dimensional (a large number of VOCs), yet only a limited number of instances is available. To be able to deal with these complicating elements, we propose an L1-regularized ordinal regression approach that is capable of exploiting multiple types of data simultaneously. Moreover, this approach allows for an automated selection of VOCs that are important to model consumer appreciation.