The novel forms of nutrition developed in food4future should be recommended and offered to consumers in an optimized form. Therefore, it is necessary to recognize the consumer’s current individual nutritional status. In this project, a type and condition asessment of consumers will be realized on the basis of sensor data and a phenotypical classification. This will be implementd initially on subjects within the framework of a nutrition study. Modern sensor technologies such as smartwatches or portable sensors provide a multitude of data. These data are going to be used in a process of machine learning to classify their wearer at a specific point in time. On the basis of physiological-medical phenotyping and correlated nutritional requirements, specific food suggestions will be given. The applied sensors record body data using non-invasive methods and should restrict the participants as little as possible. The selection of suitable sensors and the development of a suitable sensor system is also part of this project.

Out of the findings of the study, we will develop a mobile application (app) for end users, which on one hand generates food suggestions based on 'smart foods' developed by food4future and on local food offers on the other hand. In consultation with the research fields within food4future, the composition and presentation of the designed food is determined. Nutritional information of the local food is taken from commercial nutrition facts labels. All sensor data and, if necessary, additional user images and inputs will be available on a central server for further analysis.

A change desired by users, e.g. a slimmer figure or a more athletic appearance, is evaluated in a model study with previously defined objectives for the test persons. As an example, if the user desires a weight reduction, the app will then suggest a protein-rich diet. The system should then learn which forms of nutrition is successful. The target input can be supplemented by the possibility to determine person-specific preferred or rejected food. The result is a communication system that grows in time and involves the end user in the decision-making process according to his preferences.

In the later course of the project, the system should be able to independently measure the quality of its given suggestions through the establishment of feed-backed continuous learning with self-correction. To this end, the mobile measurement systems will be interconnected as part of an Internet of Things (IoT) architecture. By correlating the time course of the phenotype classification with the food recommendation, the effect of the recommended products on the test persons is included and improved during operation by algorithms of machine learning.

Text: P. Wagner, E. Gedat, TH Wildau