Page 302 - Data Science Algorithms in a Week

Artificial Intelligence for the Modeling and Prediction ... 283

The architecture can vary in terms number of internal layers, how many ANs in each
layer, the connections (fully or partially interconnected layers) between ANs and in the
transfer function of chosen for the signal processing of each AN.
From the ANN theory, it is evident, that there are many values (weights, thresholds)
which have to be set. To do so, many adaptation algorithms have been developed which
mainly fall into two basic groups: supervised and unsupervised.
Supervised algorithm requires the knowledge of the desired output. The algorithm
then calculates the output with current weights and biases. The output is compared with
targeted output and the weights and biases are adjusted by algorithm. This cycle is
repeated until the difference between targeted and calculated values is as closer as it can
get. The most applied supervised algorithms are based on gradient methods (for example
‘back propagation’) (Figure 3) and genetics (genetic algorithms). While the supervised
learning algorithm requires the knowledge of output values, the unsupervised does not
need them. It produces its own output which needs further evaluation.
When the ANN finishes the adjustments after a established number of iterations (or
epochs) it is necessary to check that it actually is ‘fit for purpose’: the prediction ability
of the network will be tested on a validating setoff data. This time only the input values
of the data will be given to the network which will calculate its own output. The
difference between the real outputs and the calculated ones can be investigated to
evaluate the prediction accuracy of the network. This can be directly visualised (as in in
Figure 4A) but eventually the performance of the predictions have to be measured by
linear correlation (see Figure 4B).

APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS TO
THE PROPERTIES AND BIOACTIVITIES OF NATURAL PRODUCTS

Two main areas of application are directly linked with the potential use of natural
products: Food industry and Pharmaceutical research. Both have started to use ANNs as a
tool to predict both the best processing methods and the final properties of the final
products made from natural sources. Perhaps ANNs are better stablished in the food
chemistry sector, whilst their use in pharmaceutical research is lagging behind.
Indeed, ANNs have been applied in almost every aspect of food science over the past
two decades, although most applications are in the development stage. ANNs are useful
tools for food safety and quality analyses, which include modeling of microbial growth
and from this predicting food safety, interpreting spectroscopic data, and predicting
physical, chemical, functional and sensory properties of various food products during
processing and distribution. (Huang, Kangas, & Rasco, 2007; Bhotmange & Shastri,
2011).