Abstract
In data analysis one could be interested in the relations among a number of data sets (data
blocks) having different origin. In food science, this can be particularly relevant. For instance,
developing a new product, one may need to understand the relation between
physical/chemical data, sensory data and consumer acceptance data. A further example could
be in process monitoring, where one of the main tasks is to figure out relations among
spectroscopic measurements on raw materials and/or during the production, process settings,
and the quality of end-product(s). Additionally, data blocks could have not only different
origin, but measurements could be taken at different time points or by multi-channel
instruments. It has been demonstrated, that it is more convenient to extract information from
multi-block data sets handling all the blocks at the same time. Namely, performing data
fusion by the means of multi-block methods. Several statistical and chemometric multi-block
methods are already available. Mainly, these are natural developments and variations of
previously widely-used methods in multivariate analysis, but the area still needs to be
explored. This PhD project is centered on method-development and method-testing in the
multi-block analysis field, with a specific focus on food analysis. Novel approaches will be
compared with other well-known methods used in the same field and they will be applied both
in regression and in classification. The new methodologies will be tested on simulations and
on real data. Attention will be also given to categorical input data (Paper IV). Additionally,
variable selection in this context will be investigated, in order to obtain reduced sub-sets,
easier to interpret (Paper II). In conclusion, due to the increasing need of handling multi-way
arrays ( i.e., structures resulting from experiments where the data are collected as a function
of more than two sources of variability), all the considerations done in the first part of the
study, will be extended to multi-way arrays (Paper III).
blocks) having different origin. In food science, this can be particularly relevant. For instance,
developing a new product, one may need to understand the relation between
physical/chemical data, sensory data and consumer acceptance data. A further example could
be in process monitoring, where one of the main tasks is to figure out relations among
spectroscopic measurements on raw materials and/or during the production, process settings,
and the quality of end-product(s). Additionally, data blocks could have not only different
origin, but measurements could be taken at different time points or by multi-channel
instruments. It has been demonstrated, that it is more convenient to extract information from
multi-block data sets handling all the blocks at the same time. Namely, performing data
fusion by the means of multi-block methods. Several statistical and chemometric multi-block
methods are already available. Mainly, these are natural developments and variations of
previously widely-used methods in multivariate analysis, but the area still needs to be
explored. This PhD project is centered on method-development and method-testing in the
multi-block analysis field, with a specific focus on food analysis. Novel approaches will be
compared with other well-known methods used in the same field and they will be applied both
in regression and in classification. The new methodologies will be tested on simulations and
on real data. Attention will be also given to categorical input data (Paper IV). Additionally,
variable selection in this context will be investigated, in order to obtain reduced sub-sets,
easier to interpret (Paper II). In conclusion, due to the increasing need of handling multi-way
arrays ( i.e., structures resulting from experiments where the data are collected as a function
of more than two sources of variability), all the considerations done in the first part of the
study, will be extended to multi-way arrays (Paper III).
| Originalsprog | Engelsk |
|---|
| Forlag | Department of Food Science, Faculty of Science, University of Copenhagen |
|---|---|
| Status | Udgivet - 2016 |